Lesson 4: 3轴码垛逆解模型及实现

对于3轴码垛机械臂控制最基本的是对其建立运动学模型,而对于3轴码垛类型机械臂来说运动学模型,其本质就是给定空间3D坐标,求解3个轴的旋转角度。

 

如上图所示,左侧为实物坐标,右侧图为抽象到坐标系的几何表示,逆解过程就是知道末端坐标,而求解各个轴的旋转角度,进而转换为步进电机的步进数。上图中涉及到5个点:

O点:坐标原点,也就是机械臂底盘原点

A点:是两个电机安装位置轴心点,此点的Z轴坐标为已知的固定量

B点:两个活动臂的连接点

C点:臂末端点

D点:末端工具安装点

E点:末端工具的工作点,逆解就是知道这点的坐标,求3个马达的选择角度

下面我们利用立体几何,和解析几何知识来进行你运算分析。

一,假设条件

  • 坐标系采用右手坐标系,如上图所示
  • 机械臂的底座位于右手坐标系的XY平面
  • 底座旋转轴的位置,即为坐标系的原点位置,即上图做所视的O点
  • 我们把AB线所表示臂称之为大臂,BC线所表示的臂称之为小臂
  • 点O,A,B,C,D,E始终处于同一平面,此平面与Y轴的夹角为α
  • OA线与AB延长线之间的夹角为β
  • AB线与CB线延长线之间的夹角为γ
  • 工具安装位置CD始终与XY平面平行
  • 工具安装点与末端工作点的直线DE始终垂直于XY平面
  • 机械臂的初始位置为A,B,C三点的Y轴坐标为O,B,C点收回到距离Z轴最近的位置
  • 初始位置α=0
  • 已知条件:OA,AB,BC,CD,DE长度
  • OA与XY平面之间的夹角θ为固定值

二、数学求解

已知E点的坐标(x,y,z)基, 于以上假设条件,求解α,β,γ。

1.求解α

c点坐标为(cx,cy,cz),与E点的坐标是简单的直角坐标系对应关系

α为整个机械臂投影到XY轴平面的投影线与X轴的角度,可以直接根据E点的坐标使用正切函数计算出α

tan(α)=y/x

α=atan(y/x)

由于在机械臂的运动范围内x坐标可能为0,所以需要对x=0的坐标点做特殊的处理,详见代码

2.求解γ

γ是三角形ABC,角ABC沿AB延长线的补角,所以只要求出角ABC,即可求出γ。角ABC只需要采用三角函数即可以求得,所以求解步骤为:

2.1 求A点的坐标(ax,ay,az)

ax=OA*cos(θ)*sin(α)

ay=OA*cos(θ)*cos(α)

az=OA*sin(θ)

2.2求解三角形AC边长

AC=sqrt((cx-ax)(cx-ax)+(cy-ay)(cy-ay)+(cz-az)(cz-az))

2.3求解γ

γ=180-acos((AB*AB+ BC*BC- AC*AC) / (2 * AB* BC))

3.求解β

三角形ABC的的BAC角是可以根据三角函数求出来,γ在OA的斜率大于AC的斜率时γ等于角BAC减去直线OA和AC的夹角

3.1.求解角ABC

ABC=acos((AB*AB+ AC*AC- BC*BC) / (2 * AB* AC))

3.2.求解OA的斜率

Koa=sin(θ) / cos(θ);

3.3.求解AC的斜率

Kac= (c_y – a_y) / (c_x – a_x);

3.4.求解OA和AC的夹角

OA_AC=atan(|Koa-Kac|/(1-Koa*Kac))

3.5.求解β

如果Koa>Kac ,β=ABC-OA_AC

如果Koa<Kac ,β=ABC+OA_AC

 

4.代码实现

逆解部分的都在planner.cpp文件的plan_buffer_line函数中

 ///////////////diego_3dof kinemitacs///////////////////////
  ///alpha  oa_map_xy to y
  float alpha = 0;
  if(y==0){///////////in the x line
    if(x>0){
      alpha=PI/2;
      c_x=x-DEFAULT_CD_LENGTH_MM;
    }else if(x<0){ alpha=-PI/2; c_x=x+DEFAULT_CD_LENGTH_MM; } }else{ float tan_alpha=x/y; if(tan_alpha==0){///in the y line if(y>0){
        alpha=0;
        c_y = y - DEFAULT_CD_LENGTH_MM;
      }else{
        alpha=PI;
        c_y = y + DEFAULT_CD_LENGTH_MM;
      }
    }else if(tan_alpha>0){
      if(y>0){
        alpha=atan(tan_alpha);
      }else{
        alpha=atan(tan_alpha)-PI;
      }
      c_y=y-DEFAULT_CD_LENGTH_MM*cos(alpha);
      c_x=x-DEFAULT_CD_LENGTH_MM*sin(alpha);
    }else{
      if(y>0){
        alpha=atan(tan_alpha);
      }else{
        alpha=PI/2-atan(tan_alpha);
      }
      c_y=y-DEFAULT_CD_LENGTH_MM*cos(alpha);
      c_x=x-DEFAULT_CD_LENGTH_MM*sin(alpha);
    }
  }

  Serial.print("------------alpha\r\n");
  Serial.print(alpha*180/PI);
  Serial.print("\r\n");
  //gamma 180- ab to bc
  float gamma = 0;
  float default_oa = DEFAULT_OA_LENGTH_MM;
  float default_ab = DEFAULT_AB_LENGTH_MM;
  float default_bc = DEFAULT_BC_LENGTH_MM;
  float default_theta = DEFAULT_THETA_DEGREE;

  float a_x = default_oa * cos(default_theta) * sin(alpha);
  float a_y = default_oa * cos(default_theta) * cos(alpha);
  float a_z = default_oa * sin(default_theta);

  float ac = sqrt(pow(c_x - a_x, 2) + pow(c_y - a_y, 2) + pow(c_z - a_z, 2));

  float cos_gamma = (pow(default_ab, 2) + pow(default_bc, 2) - pow(ac, 2)) / (2 * default_ab * default_bc);
//  Serial.print("-------------cos_gamma");
//  Serial.print(cos_gamma);

  
  //gamma = PI - acos(cos_gamma);
  gamma = acos(cos_gamma);

  Serial.print("-------------gamma\r\n");
  Serial.print(gamma*180/PI);
  Serial.print("\r\n");

  ///beta; map to the axis between oa to ab
  float beta = 0;
  float k_map_oa = sin(default_theta) / cos(default_theta);

  float a_map_x = default_oa * cos(default_theta);  ////62.47
  float a_map_y = default_oa * sin(default_theta);  ////101.7

  float c_map_x = sqrt(pow(c_x, 2) + pow(c_y, 2)); ////130
  float c_map_y = c_z; ////102
  float k_map_ac = (c_map_y - a_map_y) / (c_map_x - a_map_x);
//  Serial.print("-------------c_map_y\r\n");
//  Serial.print(c_map_y);
//  Serial.print("\r\n");
//  Serial.print("-------------a_map_y\r\n");
//  Serial.print(a_map_y);
//  Serial.print("\r\n");
//  Serial.print("-------------c_map_x\r\n");
//  Serial.print(c_map_x);
//  Serial.print("\r\n");
//  Serial.print("-------------a_map_x\r\n");
//  Serial.print(a_map_x);
//  Serial.print("\r\n");
//  Serial.print("-------------k_map_oa\r\n");
//  Serial.print(k_map_oa);
//  Serial.print("\r\n");
//  Serial.print("-------------k_map_ac\r\n");
//  Serial.print(k_map_ac);
//  Serial.print("\r\n");

  float tan_oa_ac = abs(k_map_oa - k_map_ac) / abs(1 - k_map_oa * k_map_ac);
  float cos_ab_ac = (pow(default_ab, 2) + pow(ac, 2) - pow(default_bc, 2)) / (2 * default_ab*ac);
  float ab_ac = acos(cos_ab_ac);
//  Serial.print("-------------ab_ac\r\n");
//  Serial.print(cos_ab_ac);
//  Serial.print("\r\n");
//  Serial.print(ab_ac);
//  Serial.print("\r\n");
  float ac_oa = atan(tan_oa_ac);
//  Serial.print("-------------ac_oa\r\n");
//  Serial.print(ac_oa);
//  Serial.print("\r\n");
  
  if (ac_oa < ab_ac) { if (k_map_ac > k_map_oa) {
      beta = 0 - ab_ac - atan(tan_oa_ac);
    } else {
      beta = 0 - ab_ac + atan(tan_oa_ac);
    }
  } else {
    float oc = sqrt(pow(c_map_x, 2) + pow(c_map_y, 2));
    float cos_oa_ac = (pow(default_oa, 2) + pow(ac, 2) - pow(oc, 2)) / (2 * default_oa * ac);
    float oa_ac = acos(cos_oa_ac);
    beta = PI - oa_ac - ab_ac;
  }

  Serial.print("-------------beta\r\n");
  Serial.print(beta*180/PI);
  Serial.print("\r\n");

  float degree_per_step_axis1 = 2*PI / (DEFAULT_STEPS_PER_CYCLE_AXIS_1 * DEFAULT_REDUCTION_RATE_AXIS_1 * DEFAULT_MICROSTEPS_AXIS_1); //steps per cycle
  float degree_per_step_axis2 = 2*PI / (DEFAULT_STEPS_PER_CYCLE_AXIS_2 * DEFAULT_REDUCTION_RATE_AXIS_2 * DEFAULT_MICROSTEPS_AXIS_2); //steps per cycle
  float degree_per_step_axis3 = 2*PI / (DEFAULT_STEPS_PER_CYCLE_AXIS_3 * DEFAULT_REDUCTION_RATE_AXIS_3 * DEFAULT_MICROSTEPS_AXIS_3); //steps per cycle

  float axis1_degree_steps = alpha / degree_per_step_axis1;
  float axis2_degree_steps = beta / degree_per_step_axis2;
  float axis3_degree_steps = gamma / degree_per_step_axis3;

  target[X_AXIS] = (int32_t)(axis1_degree_steps + 0.5);
  target[Y_AXIS] = (int32_t)(axis2_degree_steps + 0.5);
  target[Z_AXIS] = (int32_t)(axis3_degree_steps + 0.5); 

全部代码请见gitclub

Lesson 2: 硬件连接

一、硬件连接

Diego 3dof的硬件连接比较简单,Arduino和A4988是通过Arduino CNC sheild v3连接的,Arduino CNC sheild可以直接插在Arduino上,然后再把3个A4988插在Arduino CNC sheild上即可,连接效果如下:

CNC sheild可以支持4路步进电机,由于我们3轴码垛只使用到了3个步进电机,所以只需要插3个A4988即可,步进电机去A4988的对应关系如下图:

 

二、Arduino Pin对应关系

A4988及马达编号Arduino IO对应功能
1D2马达1步进脉冲
1D5马达1方向
2D3马达2步进脉冲
2D6马达2方向
3D4马达2步进脉冲
3D7马达2方向

 

 

三、其他

1.Diego 3轴机械臂的机械部分需要根据图纸进行钣金加工,或者也可以3D打印,钣金加工的精度要高与3D打印的精度

2. 安装过程中需要用到的不同规格的螺丝刀,电钻等工。

Lesson 26:Diego 1# —物体识别和定位

google最近公布了基于tensorflow物体识别的Api,本文将利用Diego1#的深度摄像头调用物体识别API,在识别物体的同时计算物体与出机器人摄像头的距离。原理如下:

  • Object Detection 订阅Openni发布的Image消息,识别视频帧中的物体
  • Object Depth 订阅Openni发布的Depth Image消息,根据Object Detection识别出的物体列表,对应到Depth Image的位置,计算Object的深度信息
  • Publish Image将视频帧经过处理,增加识别信息Lable后,以Compressed Image消息发布出去,可以方便其他应用订阅

1.创建diego_tensorflow包

由于我们使用的tensorflow,所以我们首先需要安装tensorflow,可以参考tensorflow官方安装说明https://www.tensorflow.org/install/install_linux

Object_detection相关依赖安装见官方安装说明https://github.com/tensorflow/models/blob/master/object_detection/g3doc/installation.md

执行如下命令创建diego_tensorflow包

catkin_create_pkg diego_tensorflow std_msgs rospy roscpp cv_bridge

在diego_tensorflow目录下创建两个子目录

  • scripts:存放相关代码
  • launch:存放launch启动文件

创建完成后diego_tensorflow目录如下图所示:

下载object_detection包:https://github.com/tensorflow/models

下载后将object_detection包上传到diego_tensorflow/scripts目录下,如果自己做模型训练还需有上传slim包到diego_tensorflow/scripts目录下

物体识别的代码都写在ObjectDetectionDemo.py文件中,其中有关识别的代码大部分参考tensorflow官方示例,这里将其包装成为一个ROS节点,并增加物体深度数据的计算

2.ROS节点源代码

#!/usr/bin/env python

import rospy
from sensor_msgs.msg import Image as ROSImage
from sensor_msgs.msg import CompressedImage as ROSImage_C
from cv_bridge import CvBridge
import cv2
import matplotlib
import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import zipfile
import uuid
from collections import defaultdict
from io import StringIO
from PIL import Image
from math import isnan

# This is needed since the notebook is stored in the object_detection folder.
from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as vis_util

class ObjectDetectionDemo():
    def __init__(self):
	rospy.init_node('object_detection_demo')
	
	# Set the shutdown function (stop the robot)
        rospy.on_shutdown(self.shutdown)
        
        self.depth_image =None
        
        self.depth_array = None
        
        model_path = rospy.get_param("~model_path", "")
        image_topic = rospy.get_param("~image_topic", "")
        depth_image_topic = rospy.get_param("~depth_image_topic", "")
        if_down=False
        self.vfc=0
        
        self._cv_bridge = CvBridge()
        
        # What model to download.
	#MODEL_NAME = 'ssd_mobilenet_v1_coco_11_06_2017'
	#MODEL_NAME='faster_rcnn_resnet101_coco_11_06_2017'
	MODEL_NAME ='ssd_inception_v2_coco_11_06_2017'
	#MODEL_NAME ='diego_object_detection_v1_07_2017'
	#MODEL_NAME ='faster_rcnn_inception_resnet_v2_atrous_coco_11_06_2017'
	MODEL_FILE = MODEL_NAME + '.tar.gz'
	DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'

	# Path to frozen detection graph. This is the actual model that is used for the object detection.
	PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

	# List of the strings that is used to add correct label for each box.
	PATH_TO_LABELS = os.path.join(model_path+'/data', 'mscoco_label_map.pbtxt')
	#PATH_TO_LABELS = os.path.join('data', 'mscoco_label_map.pbtxt')

	NUM_CLASSES = 90
	
	if if_down:
		opener = urllib.request.URLopener()
		opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
		tar_file = tarfile.open(MODEL_FILE)
		for file in tar_file.getmembers():
			file_name = os.path.basename(file.name)
			if 'frozen_inference_graph.pb' in file_name:
        			tar_file.extract(file, os.getcwd())


	rospy.loginfo("begin initilize the tf...")
	self.detection_graph = tf.Graph()
	with self.detection_graph.as_default():
		od_graph_def = tf.GraphDef()
		with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
			serialized_graph = fid.read()
			od_graph_def.ParseFromString(serialized_graph)
			tf.import_graph_def(od_graph_def, name='')

	label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
	categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
	self.category_index = label_map_util.create_category_index(categories)
	
	# Subscribe to the registered depth image
        rospy.Subscriber(depth_image_topic, ROSImage, self.convert_depth_image)
        
        # Wait for the depth image to become available
        #rospy.wait_for_message('depth_image', ROSImage)
	
	self._sub = rospy.Subscriber(image_topic, ROSImage, self.callback, queue_size=1)	
	self._pub = rospy.Publisher('object_detection', ROSImage_C, queue_size=1)
	
	rospy.loginfo("initialization has finished...")
	
	
    def convert_depth_image(self, ros_image):
        # Use cv_bridge() to convert the ROS image to OpenCV format
        # The depth image is a single-channel float32 image
        self.depth_image = self._cv_bridge.imgmsg_to_cv2(ros_image, "32FC1")

        # Convert the depth image to a Numpy array
        self.depth_array = np.array(self.depth_image, dtype=np.float32)
        #print(self.depth_array)
        
    def callback(self,image_msg):
	if self.vfc<12:
		self.vfc=self.vfc+1
	else:
		self.callbackfun(image_msg)
		self.vfc=0	
		    	
    def box_depth(self,boxes,im_width, im_height):
	# Now compute the depth component
        depth=[]
	for row in boxes[0]:
		n_z = sum_z = mean_z = 0
		# Get the min/max x and y values from the ROI
		if row[0]<row[1]:
			min_x = row[0]*im_width
			max_x = row[1]*im_width
		else:
			min_x = row[1]*im_width
			max_x = row[0]*im_width
			
		if row[2]<row[3]:
			min_y = row[2]*im_height
			max_y = row[3]*im_height
		else:
			min_y = row[3]*im_height
			max_y = row[2]*im_height
		# Get the average depth value over the ROI
		for x in range(int(min_x), int(max_x)):
            		for y in range(int(min_y), int(max_y)):
                		try:
					z = self.depth_array[y, x]
				except:
					continue
                
				# Depth values can be NaN which should be ignored
				if isnan(z):
					z=6
					continue
				else:
					sum_z = sum_z + z
					n_z += 1 
			mean_z = sum_z / (n_z+0.01)
		depth.append(mean_z)
	return depth
    def callbackfun(self, image_msg):
	with self.detection_graph.as_default():
		with tf.Session(graph=self.detection_graph) as sess:
			 cv_image = self._cv_bridge.imgmsg_to_cv2(image_msg, "bgr8")
			 #cv_image = (self._cv_bridge.imgmsg_to_cv2(image_msg, "bgr8"))[300:450, 150:380]
			 pil_img = Image.fromarray(cv_image)			 
			 (im_width, im_height) = pil_img.size			 
			 # the array based representation of the image will be used later in order to prepare the
			 # result image with boxes and labels on it.
			 image_np =np.array(pil_img.getdata()).reshape((im_height, im_width, 3)).astype(np.uint8)
			 # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
			 image_np_expanded = np.expand_dims(image_np, axis=0)
			 image_tensor = self.detection_graph.get_tensor_by_name('image_tensor:0')
			 # Each box represents a part of the image where a particular object was detected.
			 boxes = self.detection_graph.get_tensor_by_name('detection_boxes:0')
			 # Each score represent how level of confidence for each of the objects.
			 # Score is shown on the result image, together with the class label.
			 scores = self.detection_graph.get_tensor_by_name('detection_scores:0')
			 classes = self.detection_graph.get_tensor_by_name('detection_classes:0')
			 num_detections = self.detection_graph.get_tensor_by_name('num_detections:0')
			
			 # Actual detection.
			 (boxes, scores, classes, num_detections) = sess.run(
			 	[boxes, scores, classes, num_detections],
			 	feed_dict={image_tensor: image_np_expanded})
			 box_depths=self.box_depth(boxes,im_width,im_height)
			 print(box_depths)
			 # Visualization of the results of a detection.
			 vis_util.visualize_boxes_and_labels_on_image_array(
			 	image_np,
			 	np.squeeze(boxes),
			 	np.squeeze(classes).astype(np.int32),
			 	np.squeeze(scores),
			 	self.category_index,
			 	use_normalized_coordinates=True,
			 	line_thickness=8)
			 
			 ros_compressed_image=self._cv_bridge.cv2_to_compressed_imgmsg(image_np)
			 self._pub.publish(ros_compressed_image)
			
    
    def shutdown(self):
        rospy.loginfo("Stopping the tensorflow object detection...")
        rospy.sleep(1) 
        
if __name__ == '__main__':
    try:
        ObjectDetectionDemo()
        rospy.spin()
    except rospy.ROSInterruptException:
        rospy.loginfo("RosTensorFlow_ObjectDetectionDemo has started.")

下面我们来解释主要的代码逻辑

    def __init__(self):
	rospy.init_node('object_detection_demo')
	
	# Set the shutdown function (stop the robot)
        rospy.on_shutdown(self.shutdown)
        
        self.depth_image =None
        
        self.depth_array = None
        
        model_path = rospy.get_param("~model_path", "")
        image_topic = rospy.get_param("~image_topic", "")
        depth_image_topic = rospy.get_param("~depth_image_topic", "")
        if_down=False
        self.vfc=0
        
        self._cv_bridge = CvBridge()

以上代码是ROS的标准初始化代码,变量的初始化,及launch文件中参数的读取

  • model_path定义object_detection所使用的模型路径
  • image_topic订阅的image主题
  • depth_image_topic订阅的深度image主题
        # What model to download.
	#MODEL_NAME = 'ssd_mobilenet_v1_coco_11_06_2017'
	#MODEL_NAME='faster_rcnn_resnet101_coco_11_06_2017'
	MODEL_NAME ='ssd_inception_v2_coco_11_06_2017'
	#MODEL_NAME ='diego_object_detection_v1_07_2017'
	#MODEL_NAME ='faster_rcnn_inception_resnet_v2_atrous_coco_11_06_2017'
	MODEL_FILE = MODEL_NAME + '.tar.gz'
	DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'

	# Path to frozen detection graph. This is the actual model that is used for the object detection.
	PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

	# List of the strings that is used to add correct label for each box.
	PATH_TO_LABELS = os.path.join(model_path+'/data', 'mscoco_label_map.pbtxt')
	#PATH_TO_LABELS = os.path.join('data', 'mscoco_label_map.pbtxt')

	NUM_CLASSES = 90
	
	if if_down:
		opener = urllib.request.URLopener()
		opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
		tar_file = tarfile.open(MODEL_FILE)
		for file in tar_file.getmembers():
			file_name = os.path.basename(file.name)
			if 'frozen_inference_graph.pb' in file_name:
        			tar_file.extract(file, os.getcwd())

以上代码设置object_detection所使用的模型,及下载解压相应的文件,这里设置了一个if_down的开关,第一次运行的时候可以打开此开关下载,以后可以关掉,因为下载的时间比较长,下载一次后面就无需再下载了。

self.detection_graph = tf.Graph()
	with self.detection_graph.as_default():
		od_graph_def = tf.GraphDef()
		with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
			serialized_graph = fid.read()
			od_graph_def.ParseFromString(serialized_graph)
			tf.import_graph_def(od_graph_def, name='')

	label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
	categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
	self.category_index = label_map_util.create_category_index(categories)

以上代码是tensorflow的初始化代码。

        # Subscribe to the registered depth image
        rospy.Subscriber(depth_image_topic, ROSImage, self.convert_depth_image)
        
        # Wait for the depth image to become available
        #rospy.wait_for_message('depth_image', ROSImage)
	
	self._sub = rospy.Subscriber(image_topic, ROSImage, self.callback, queue_size=1)	
	self._pub = rospy.Publisher('object_detection', ROSImage_C, queue_size=1)

以上代码,我们定义此节点订阅depth_image和image两个topic,同时发布一个名为object_detection的Compressed Imagetopic
depth_image的回调函数是convert_depth_image
image的回调函数是callback

    def convert_depth_image(self, ros_image):
        # Use cv_bridge() to convert the ROS image to OpenCV format
        # The depth image is a single-channel float32 image
        self.depth_image = self._cv_bridge.imgmsg_to_cv2(ros_image, "32FC1")

        # Convert the depth image to a Numpy array
        self.depth_array = np.array(self.depth_image, dtype=np.float32)
        #print(self.depth_array)

以上是depth_image处理的回调函数,首先将depth_image主题转换成opencv类型的,然后在将图片转换为numpy数组,赋值给depth_array成员变量

    def callback(self,image_msg):
	if self.vfc<12:
		self.vfc=self.vfc+1
	else:
		self.callbackfun(image_msg)
		self.vfc=0

以上是image处理的回调函数,这里控制视频帧的处理频率,主要是为了减少运算量,可以灵活调整,最终视频帧的处理是在callbackfun中处理的

    def box_depth(self,boxes,im_width, im_height):
	# Now compute the depth component
        depth=[]
	for row in boxes[0]:
		n_z = sum_z = mean_z = 0
		# Get the min/max x and y values from the ROI
		if row[0]<row[1]:
			min_x = row[0]*im_width
			max_x = row[1]*im_width
		else:
			min_x = row[1]*im_width
			max_x = row[0]*im_width
			
		if row[2]<row[3]:
			min_y = row[2]*im_height
			max_y = row[3]*im_height
		else:
			min_y = row[3]*im_height
			max_y = row[2]*im_height
		# Get the average depth value over the ROI
		for x in range(int(min_x), int(max_x)):
            		for y in range(int(min_y), int(max_y)):
                		try:
					z = self.depth_array[y, x]
				except:
					continue
                
				# Depth values can be NaN which should be ignored
				if isnan(z):
					z=6
					continue
				else:
					sum_z = sum_z + z
					n_z += 1 
			mean_z = sum_z / (n_z+0.01)
		depth.append(mean_z)
	return depth

以上代码是深度数据计算,输入参数boxes就是object_detection识别出来的物体的矩形标识rect,我们根据物体的矩形范围,匹配深度图片相应的区域,计算区域内的平均深度值作为此物体的深度数据。返回一个与boxes想对应的1维数组
由于深度图,和一般图片是异步处理,可能出现帧不对应的问题,在这里处理的比较简单,没有考虑此问题,可以通过缓存深度图片的方式来解决,通过时间戳来匹配最近的深度图片。

def callbackfun(self, image_msg):
	with self.detection_graph.as_default():
		with tf.Session(graph=self.detection_graph) as sess:
			 cv_image = self._cv_bridge.imgmsg_to_cv2(image_msg, "bgr8")
			 #cv_image = (self._cv_bridge.imgmsg_to_cv2(image_msg, "bgr8"))[300:450, 150:380]
			 pil_img = Image.fromarray(cv_image)			 
			 (im_width, im_height) = pil_img.size			 
			 # the array based representation of the image will be used later in order to prepare the
			 # result image with boxes and labels on it.
			 image_np =np.array(pil_img.getdata()).reshape((im_height, im_width, 3)).astype(np.uint8)
			 # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
			 image_np_expanded = np.expand_dims(image_np, axis=0)
			 image_tensor = self.detection_graph.get_tensor_by_name('image_tensor:0')
			 # Each box represents a part of the image where a particular object was detected.
			 boxes = self.detection_graph.get_tensor_by_name('detection_boxes:0')
			 # Each score represent how level of confidence for each of the objects.
			 # Score is shown on the result image, together with the class label.
			 scores = self.detection_graph.get_tensor_by_name('detection_scores:0')
			 classes = self.detection_graph.get_tensor_by_name('detection_classes:0')
			 num_detections = self.detection_graph.get_tensor_by_name('num_detections:0')
			
			 # Actual detection.
			 (boxes, scores, classes, num_detections) = sess.run(
			 	[boxes, scores, classes, num_detections],
			 	feed_dict={image_tensor: image_np_expanded})
			 box_depths=self.box_depth(boxes,im_width,im_height)
			 # Visualization of the results of a detection.
			 vis_util.visualize_boxes_and_labels_on_image_array(
			 	image_np,
			 	np.squeeze(boxes),
			 	np.squeeze(classes).astype(np.int32),
			 	np.squeeze(scores),
			 	self.category_index,
                                box_depths,
			 	use_normalized_coordinates=True,
			 	line_thickness=8)
			 
			 ros_compressed_image=self._cv_bridge.cv2_to_compressed_imgmsg(image_np)
			 self._pub.publish(ros_compressed_image)

以上代码是图片的回调函数,主要是将image消息转换为opencv格式,然后再转换成numpy数组,调用object_detection来识别图片中的物体,再调用 vis_util.visualize_boxes_and_labels_on_image_array将识别出来的物体在图片上标识出来,最后将处理后的图片发布为compressed_image类型的消息

现在我们只需要简单修改一下Object_detection/utils目录下的visualization_utils.py文件,就可以显示深度信息

def visualize_boxes_and_labels_on_image_array(image,
                                              boxes,
                                              classes,
                                              scores,
                                              category_index,
	                                      box_depths=None,
                                              instance_masks=None,
                                              keypoints=None,
                                              use_normalized_coordinates=False,
                                              max_boxes_to_draw=20,
                                              min_score_thresh=.5,
                                              agnostic_mode=False,
                                              line_thickness=4):

在visualize_boxes_and_labels_on_image_array定义中增加box_depths=None,缺省值为None

  for i in range(min(max_boxes_to_draw, boxes.shape[0])):
    if scores is None or scores[i] > min_score_thresh:
      box = tuple(boxes[i].tolist())
      if instance_masks is not None:
        box_to_instance_masks_map[box] = instance_masks[i]
      if keypoints is not None:
        box_to_keypoints_map[box].extend(keypoints[i])
      if scores is None:
        box_to_color_map[box] = 'black'
      else:
        if not agnostic_mode:
          if classes[i] in category_index.keys():
            class_name = category_index[classes[i]]['name']
          else:
            class_name = 'N/A'
          display_str = '{}: {}%'.format(
              class_name,
              int(100*scores[i]))
        else:
          display_str = 'score: {}%'.format(int(100 * scores[i]))
          
        #modify by diego robot
        if box_depths!=None:
        	display_str=display_str+"\ndepth: "+str(box_depths[i])
        	global depth_info
        	depth_info=True
        else:
        	global depth_info
        	depth_info=False
        #######################
        box_to_display_str_map[box].append(display_str)
        if agnostic_mode:
          box_to_color_map[box] = 'DarkOrange'
        else:
          box_to_color_map[box] = STANDARD_COLORS[
              classes[i] % len(STANDARD_COLORS)]

在第一个for循环里面,的box_to_display_str_map[box].append(display_str)一句前面增加如上diego robot修改部分代码

depth_info=False

在文件的开头部分定义全局变量,表示是否有深度信息传递进来

 # Reverse list and print from bottom to top.
  for display_str in display_str_list[::-1]: 
    text_width, text_height = font.getsize(display_str)    
    #modify by william
    global depth_info
    if depth_info:
  	text_height=text_height*2
    ###################
    	
    margin = np.ceil(0.05 * text_height)
    draw.rectangle(
        [(left, text_bottom - text_height - 2 * margin), (left + text_width,
                                                          text_bottom)],
        fill=color)
    draw.text(
        (left + margin, text_bottom - text_height - margin),
        display_str,
        fill='black',
        font=font)
    text_bottom -= text_height - 2 * margin

在draw_bounding_box_on_image函数的margin = np.ceil(0.05 * text_height)一句前面增加如上diego robot修改部分代码

3.launch文件

<launch>
   <node pkg="diego_tensorflow" name="ObjectDetectionDemo" type="ObjectDetectionDemo.py" output="screen">

       <param name="image_topic" value="/camera/rgb/image_raw" />

       <param name="depth_image_topic" value="/camera/depth_registered/image" />

       <param name="model_path" value="$(find diego_tensorflow)/scripts/object_detection" />

   </node>

</launch>

launch文件中定义了相应的参数,image_topic,depth_image_topic,model_path,读者可以根据自己的实际情况设定

4.启动节点

启动openni

roslaunch diego_vision openni_node.launch 

启动object_detection

roslaunch diego_tensorflow object_detection_demo.launch

5.通过手机APP订阅object_detection

我们只需要设置一个image_topic为object_detection,既可以在手机上看到物体识别的效果

object_detection物体识别是建立在训练模型上的,本文所采用的是官方所提供的训练好的模型,识别率还是比较高的。但是如果机器人所工作的环境与训练的图片差异比较大,识别率会大大降低。所以我们在实际使用中可以训练自己物体模型,在结合机器人的其他传感器和装置来达到实际应用的要求。比如可以应用在人员流量的监控,车辆的监控,或者特定物体的跟踪,再配合机械臂,深度相机,可以实现物体位置的判断,抓取等功能。

Lesson 25:Diego 1# 4WD —3:上位机通讯

ROS Arduino Bridge本质上是上位机通过串口发送控制命令来实现对Arduino的控制,所以我们要实现4驱的控制,我们也必须的修改通讯部分

1.Arduino firmware修改

1.1 command.h

在此文件中增加4驱所需的命令及宏定义

#ifndef COMMANDS_H
#define COMMANDS_H

#define ANALOG_READ    'a'
#define GET_BAUDRATE   'b'
#define PIN_MODE       'c'
#define DIGITAL_READ   'd'
#define READ_ENCODERS  'e'
#define MOTOR_SPEEDS   'm'
#define PING           'p'
#define RESET_ENCODERS 'r'
#define SERVO_WRITE    's'
#define SERVO_READ     't'
#define UPDATE_PID     'u'
#define DIGITAL_WRITE  'w'
#define ANALOG_WRITE   'x'
#define LEFT            0
#define RIGHT           1
#define LEFT_H          2 //新增
#define RIGHT_H         3 //新增
#define READ_PIDOUT    'f'
#define READ_PIDIN     'i'
#define READ_MPU6050   'g'

#endif

1.2 RosArduinoBridge-diego.ino

此文件是主程序,由于此文件代码比较多,故这里只介绍新增部分代码,完整的代码请见github

在runCommand()函数中修改在4驱模式下读取pidin 的代码

    case READ_PIDIN:
      Serial.print(readPidIn(LEFT));
      Serial.print(" ");
#ifdef L298P      
      Serial.println(readPidIn(RIGHT));
#endif
#ifdef L298P_4WD 
      Serial.print(readPidIn(RIGHT));
      Serial.print(" ");
      Serial.print(readPidIn(LEFT_H));
      Serial.print(" ");
      Serial.println(readPidIn(RIGHT_H));
#endif      
      break;

在runCommand()函数中修改在4驱模式下读取pidout 的代码

    case READ_PIDOUT:
      Serial.print(readPidOut(LEFT));
      Serial.print(" ");
#ifdef L298P      
      Serial.println(readPidOut(RIGHT));
#endif
#ifdef L298P_4WD 
      Serial.print(readPidOut(RIGHT));
      Serial.print(" ");
      Serial.print(readPidOut(LEFT_H));
      Serial.print(" ");
      Serial.println(readPidOut(RIGHT_H));
#endif       
      break;

在runCommand()函数中修改在4驱模式下读取编码器数据的代码

    case READ_ENCODERS:
      Serial.print(readEncoder(LEFT));
      Serial.print(" ");
#ifdef L298P 
      Serial.println(readEncoder(RIGHT));
#endif
#ifdef L298P_4WD
      Serial.print(readEncoder(RIGHT));
      Serial.print(" ");
      Serial.print(readEncoder(LEFT_H));
      Serial.print(" ");
      Serial.println(readEncoder(RIGHT_H));
#endif      
      break;

在runCommand()函数中修改在4驱模式下设置马达速度的代码

    case MOTOR_SPEEDS:
      /* Reset the auto stop timer */
      lastMotorCommand = millis();
      if (arg1 == 0 && arg2 == 0) {
#ifdef L298P        
        setMotorSpeeds(0, 0);
#endif

#ifdef L298P_4WD
        setMotorSpeeds(0, 0, 0, 0);
#endif        
        moving = 0;
      }
      else moving = 1;
      leftPID.TargetTicksPerFrame = arg1;
      rightPID.TargetTicksPerFrame = arg2;
#ifdef L298P_4WD      
      leftPID_h.TargetTicksPerFrame = arg1;
      rightPID_h.TargetTicksPerFrame = arg2;
#endif       
      Serial.println("OK");
      break;

在runCommand()函数中修改在4驱模式下更新PID参数的代码

    case UPDATE_PID:
      while ((str = strtok_r(p, ":", &p)) != '\0') {
        pid_args[i] = atoi(str);
        i++;
      }

      left_Kp = pid_args[0];
      left_Kd = pid_args[1];
      left_Ki = pid_args[2];
      left_Ko = pid_args[3];

      right_Kp = pid_args[4];
      right_Kd = pid_args[5];
      right_Ki = pid_args[6];
      right_Ko = pid_args[7];

#ifdef L298P_4WD

      left_h_Kp = pid_args[0];
      left_h_Kd = pid_args[1];
      left_h_Ki = pid_args[2];
      left_h_Ko = pid_args[3];

      right_h_Kp = pid_args[4];
      right_h_Kd = pid_args[5];
      right_h_Ki = pid_args[6];
      right_h_Ko = pid_args[7];
#endif
      
      Serial.println("OK");
      break;

在setup()函数中设置在4驱模式下新增的pin对应的寄存器的操作

void setup() {
  Serial.begin(BAUDRATE);
#ifdef USE_BASE
#ifdef ARDUINO_ENC_COUNTER
  //set as inputs
  DDRD &= ~(1 << LEFT_ENC_PIN_A);
  DDRD &= ~(1 << LEFT_ENC_PIN_B);
  DDRC &= ~(1 << RIGHT_ENC_PIN_A);
  DDRC &= ~(1 << RIGHT_ENC_PIN_B);

#ifdef L298P_4WD
  DDRD &= ~(1 << LEFT_H_ENC_PIN_A);
  DDRD &= ~(1 << LEFT_H_ENC_PIN_B);
  DDRC &= ~(1 << RIGHT_H_ENC_PIN_A);
  DDRC &= ~(1 << RIGHT_H_ENC_PIN_B);
#endif  

  //enable pull up resistors
  PORTD |= (1 << LEFT_ENC_PIN_A);
  PORTD |= (1 << LEFT_ENC_PIN_B);
  PORTC |= (1 << RIGHT_ENC_PIN_A);
  PORTC |= (1 << RIGHT_ENC_PIN_B);

#ifdef L298P_4WD
  PORTD &= ~(1 << LEFT_H_ENC_PIN_A);
  PORTD &= ~(1 << LEFT_H_ENC_PIN_B);
  PORTC &= ~(1 << RIGHT_H_ENC_PIN_A);
  PORTC &= ~(1 << RIGHT_H_ENC_PIN_B);
#endif    
  // tell pin change mask to listen to left encoder pins
  PCMSK2 |= (1 << LEFT_ENC_PIN_A) | (1 << LEFT_ENC_PIN_B);
  // tell pin change mask to listen to right encoder pins
  PCMSK1 |= (1 << RIGHT_ENC_PIN_A) | (1 << RIGHT_ENC_PIN_B);

#ifdef L298P_4WD
  // tell pin change mask to listen to left encoder pins
  PCMSK2 |= (1 << LEFT_ENC_PIN_A) | (1 << LEFT_ENC_PIN_B) | (1 << LEFT_H_ENC_PIN_A) | (1 << LEFT_H_ENC_PIN_B);
  // tell pin change mask to listen to right encoder pins
  PCMSK1 |= (1 << RIGHT_ENC_PIN_A) | (1 << RIGHT_ENC_PIN_B) | (1 << RIGHT_H_ENC_PIN_A) | (1 << RIGHT_H_ENC_PIN_B);
#endif
  // enable PCINT1 and PCINT2 interrupt in the general interrupt mask
  PCICR |= (1 << PCIE1) | (1 << PCIE2);
#endif
  initMotorController();
  resetPID();
#endif

  /* Attach servos if used */
#ifdef USE_SERVOS
  int i;
  for (i = 0; i < N_SERVOS; i++) {
    servosPos[i]=90;
  }
  servodriver.begin();
  servodriver.setPWMFreq(50);
#endif
}

在loop()函数中修改在4驱模式下自动停止的逻辑

  // Check to see if we have exceeded the auto-stop interval
  if ((millis() - lastMotorCommand) > AUTO_STOP_INTERVAL) {
    ;
#ifdef L298P    
    setMotorSpeeds(0, 0);
#endif
#ifdef L298P_4WD
    setMotorSpeeds(0, 0, 0, 0);
#endif
    moving = 0;
  }

#endif

2.上位机代码修改

上位机的代码修改主要是针对arduino_driver.py和base_controller.py的修改。

2.1 arduino_driver.py修改

修改def get_pidin(self):使其支持4个电机的pidin读取

    def get_pidin(self):
        values = self.execute_array('i')
        print("pidin_raw_data: "+str(values))
        if len(values) not in [2,4]:
            print "pidin was not 2 or 4 for 4wd"
            raise SerialException
            return None
        else:                                                                  
            return values

修改def get_pidout(self):使其支持4个电机的pidout读取

    def get_pidout(self):
        values = self.execute_array('f')
        print("pidout_raw_data: "+str(values))
        if len(values) not in [2,4]:
            print "pidout was not 2 or 4 for 4wd"
            raise SerialException
            return None
        else:                                                                  
            return values

修改def get_encoder_counts(self):使其支持4个电机的编码器数据读取

    def get_encoder_counts(self):
        values = self.execute_array('e')
        if len(values) not in [2,4]:
            print "Encoder count was not 2 or 4 for 4wd"
            raise SerialException
            return None
        else:

修改完后,将arduino的固件更新,就可以启动4驱底盘控制了,要注意的时候要同时打开上位机和arduino上的4驱开关,如果是使用2驱的代码,也要记得关掉4驱的开关。

3.启动4驱底盘控制

现在执行如下命令可以启动4驱底盘控制

roslaunch diego_nav diego_arduino_run.launch 

现在我们可以通过配套的ROS APP连接Diego#进行控制

首先创建一个新的ROS机器人连接,设置相应的ROS Master ip,和cmd_vel主题。

连接上机器人后,可以选择操作杆,和重力感应来对机器人操作

Lesson 24:Diego 1# 4驱底盘 —2:马达控制

针对马达控制我们主要需要修改两部分:

  • 驱动部分,使其可以支持同时控制4个马达
  • PID调速部分,试验表明,虽然是一样的型号的马达,但是给相同的PWM值,马达的转速也不一样,所以我们需要分别对4个马达进行PID调速

1.马达驱动

1.1. DualL298PMotorShield4WD.h文件的修改
增加了4个马达对应的PWM和方向控制的Pin引脚定义,和4个马达速度的设置函数

#ifndef DualL298PMotorShield4WD_h
#define DualL298PMotorShield4WD_h

#include 

class DualL298PMotorShield4WD
{
  public:  
    // CONSTRUCTORS
    DualL298PMotorShield4WD(); // Default pin selection.
    
    // PUBLIC METHODS
    void init(); // Initialize TIMER 1, set the PWM to 20kHZ. 
    void setM1Speed(int speed); // Set speed for M1.
    void setM2Speed(int speed); // Set speed for M2.
    void setM3Speed(int speed); // Set speed for M3.
    void setM4Speed(int speed); // Set speed for M4.
    void setSpeeds(int m1Speed, int m2Speed, int m3Speed, int m4Speed); // Set speed for both M1 and M2.
    
  private:
  
    // left motor
    static const unsigned char _M1DIR = 12;
    static const unsigned char _M2DIR = 7;
    static const unsigned char _M1PWM = 10;
    static const unsigned char _M2PWM = 6;
    
    // right motor
    static const unsigned char _M4DIR = 8;
    static const unsigned char _M3DIR = 13;
    static const unsigned char _M4PWM = 9;
    static const unsigned char _M3PWM = 11;
    
};

#endif

1.2. DualL298PMotorShield4WD.cpp文件的修改

主要是针对4个马达速度设置函数的实现,逻辑都是一样的,这里只截取一个马达的代码,代码的逻辑请看代码注释

// Set speed for motor 4, speed is a number betwenn -400 and 400
void DualL298PMotorShield4WD::setM4Speed(int speed)
{
  unsigned char reverse = 0;
  
  if (speed < 0) //速度是否小于0 { speed = -speed; // 如果小于0,则是反转 reverse = 1; // 反转标志变量设置为1 } if (speed > 255)  // 限定最大速度为255
    speed = 255;
  if (reverse)  //反转状态下
  {
    digitalWrite(_M4DIR,LOW);//设定马达转向的pin位低电平
    analogWrite(_M4PWM, speed);
  }
  else //正向转动状态下
  {
    digitalWrite(_M4DIR,HIGH);设定马达转向的pin位高电平
    analogWrite(_M4PWM, speed);
  }
}

在setSpeeds函数中分别调用4个马达的速度设置函数。

// Set speed for motor 1, 2, 3, 4
void DualL298PMotorShield4WD::setSpeeds(int m1Speed, int m2Speed, int m3Speed, int m4Speed)
{
  setM1Speed(m1Speed);
  setM2Speed(m2Speed);
  setM3Speed(m3Speed);
  setM4Speed(m4Speed);  
}

DualL298PMotorShield4WD.h和DualL298PMotorShield4WD.cpp修改完成后,在arduino 的library目录下新建一个名为dual-L298P-motor-shield-master-4wd的目录,将两个文件放到此文件夹下

现在我们打开android ide的library就可以看到我们刚才添加的库

1.1.3 motor_driver.h的修改

增加4驱马达控制的函数setMotorSpeeds,参数为4个马达的速度

void initMotorController();
void setMotorSpeed(int i, int spd);
#ifdef L298P
void setMotorSpeeds(int leftSpeed, int rightSpeed);
#endif
#ifdef L298P_4WD
void setMotorSpeeds(int leftSpeed_1, int leftSpeed_2, int rightSpeed_1, int rightSpeed_2);
#endif

1.1.4 motor_driver.ino的修改

#ifdef L298P_4WD
// A convenience function for setting both motor speeds
void setMotorSpeeds(int leftSpeed_1, int leftSpeed_2, int rightSpeed_1, int rightSpeed_2){
  setMotorSpeed(1, leftSpeed_1);
  setMotorSpeed(3, rightSpeed_1);
  setMotorSpeed(2, leftSpeed_2);
  setMotorSpeed(4, rightSpeed_2);
}
#endif
#else
#error A motor driver must be selected!
#endif

2.PID控制

PID控制都在diff_controller.h文件中定义修改

新增两个在4驱模式下的PID控制变量

#ifdef L298P_4WD
SetPointInfo leftPID_h, rightPID_h;
#endif

新增两个在4驱模式下的PID控制参数

#ifdef L298P_4WD

int left_h_Kp=Kp;
int left_h_Kd=Kd;
int left_h_Ki=Ki;
int left_h_Ko=Ko;

int right_h_Kp=Kp;
int right_h_Kd=Kd;
int right_h_Ki=Ki;
int right_h_Ko=Ko;

#endif

新增两个在4驱模式下的PID控制函数

#ifdef L298P_4WD
/* PID routine to compute the next motor commands */
void dorightPID_h(SetPointInfo * p) {
  long Perror;
  long output;
  int input;

  //Perror = p->TargetTicksPerFrame - (p->Encoder - p->PrevEnc);
  input =  p->Encoder-p->PrevEnc ;
  Perror = p->TargetTicksPerFrame - input;

  /*
  * Avoid derivative kick and allow tuning changes,
  * see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-derivative-kick/
  * see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-tuning-changes/
  */
  //output = (Kp * Perror + Kd * (Perror - p->PrevErr) + Ki * p->Ierror) / Ko;
  // p->PrevErr = Perror;
  output = (right_h_Kp * Perror - right_h_Kd * (input - p->PrevInput) + p->ITerm) / right_h_Ko;
  p->PrevEnc = p->Encoder;

  output += p->output;
  // Accumulate Integral error *or* Limit output.
  // Stop accumulating when output saturates
  if (output >= MAX_PWM)
    output = MAX_PWM;
  else if (output <= -MAX_PWM) output = -MAX_PWM; else /* * allow turning changes, see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-tuning-changes/ */ p->ITerm += left_h_Ki * Perror;

  p->output = output;
  p->PrevInput = input;
//  Serial.println("right output:");
//  Serial.println(p->output);
}
#endif
...
#ifdef L298P_4WD

/* PID routine to compute the next motor commands */
void doleftPID_h(SetPointInfo * p) {
  long Perror;
  long output;
  int input;

  //Perror = p->TargetTicksPerFrame - (p->Encoder - p->PrevEnc);
  input = p->Encoder-p->PrevEnc ;
  Perror = p->TargetTicksPerFrame - input;

  /*
  * Avoid derivative kick and allow tuning changes,
  * see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-derivative-kick/
  * see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-tuning-changes/
  */
  //output = (Kp * Perror + Kd * (Perror - p->PrevErr) + Ki * p->Ierror) / Ko;
  // p->PrevErr = Perror;
  output = (left_h_Kp * Perror - left_h_Kd * (input - p->PrevInput) + p->ITerm) / left_h_Ko;
  p->PrevEnc = p->Encoder;

  output += p->output;
  // Accumulate Integral error *or* Limit output.
  // Stop accumulating when output saturates
  if (output >= MAX_PWM)
    output = MAX_PWM;
  else if (output <= -MAX_PWM) output = -MAX_PWM; else /* * allow turning changes, see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-tuning-changes/ */ p->ITerm += left_h_Ki * Perror;

  p->output = output;
  p->PrevInput = input;
//  Serial.println("left output:");
//  Serial.println(p->output);
}
#endif

修改readPIDIn函数,增加在4驱模式下,pidin的读取

long readPidIn(int i) {
  long pidin=0;
  if (i == LEFT){
    pidin = leftPID.PrevInput;
  }else if (i == RIGHT){
    pidin = rightPID.PrevInput;
  }
#ifdef L298P_4WD
  else if (i== RIGHT_H){
    pidin = rightPID_h.PrevInput;
  }else{
    pidin = leftPID_h.PrevInput;
  }
#endif  
  return pidin;
}

修改readPIDOut函数,增加在4驱模式下,pidOut的读取

long readPidOut(int i) {
  long pidout=0;
  if (i == LEFT){
    pidout = leftPID.output;
  }else if (i == RIGHT){
    pidout = rightPID.output;
  }
#ifdef L298P_4WD
  else if (i == RIGHT_H){
    pidout = rightPID_h.output;
  }else{
    pidout = leftPID_h.output;
  }
#endif   
  return pidout;
}

修改updatePID函数,增加在4驱模式下马达PID控制的调用

void updatePID() {
  /* Read the encoders */
  leftPID.Encoder =readEncoder(LEFT);
  rightPID.Encoder =readEncoder(RIGHT);
#ifdef L298P_4WD 
  leftPID_h.Encoder =readEncoder(LEFT_H);
  rightPID_h.Encoder =readEncoder(RIGHT_H);
#endif
  
  /* If we're not moving there is nothing more to do */
  if (!moving){
    /*
    * Reset PIDs once, to prevent startup spikes,
    * see http://brettbeauregard.com/blog/2011/04/improving-the-beginner%E2%80%99s-pid-initialization/
    * PrevInput is considered a good proxy to detect
    * whether reset has already happened
    */
#ifdef L298P    
    if (leftPID.PrevInput != 0 || rightPID.PrevInput != 0) resetPID();
#endif
#ifdef L298P_4WD
    if (leftPID.PrevInput != 0 || rightPID.PrevInput != 0 || leftPID_h.PrevInput != 0 || rightPID_h.PrevInput != 0) resetPID();
#endif    
    return;
  }

  /* Compute PID update for each motor */
  dorightPID(&rightPID);
  doleftPID(&leftPID);
#ifdef L298P_4WD
  dorightPID_h(&rightPID_h);
  doleftPID_h(&leftPID_h);
#endif  

  /* Set the motor speeds accordingly */
#ifdef L298P   
  setMotorSpeeds(leftPID.output, rightPID.output);
#endif

#ifdef L298P_4WD
  setMotorSpeeds(leftPID.output,leftPID_h.output, rightPID.output,rightPID_h.output);
#endif

}

至此4个马达已经可以独立驱动,独立的PID调速了,在下一篇教程中,会介绍如和上位机互动起来,让4驱底盘跑起来。

Lesson 23:Diego 1# 4驱底盘 —1:编码器

4驱底盘由于四个轮子可以独立控制,所以具有优秀的通过性,这篇文章介绍Diego 1#的四驱底盘,所有源代码都已经上传到github。

这里会分几篇文章来介绍4驱动版diego 1#的开发,这篇我们主要说明4驱动底盘4个马达编码器数据的读取。

1.1硬件说明

  • 底盘材质:铝合金材质
  • 轮胎:12cm 橡胶轮胎
  • 电机:370直流电机,带霍尔码盘测速,输出AB项编码信号

1.2 控制器

  • 主控制器 arduino UNO,使用uno分别对4个电机进行方向,PWM控制,并采用终端方式采集4个电机输出的编码信号
  • 电机控制器 两块L298p, 网上采购,控制引脚不同
  • 上位机:树莓派,或者mini pc

2.编码器数据读取

所以代码都基于diego 1# github上的代码进行修改,这里只说明4驱版本部分的代码

首先我们在ROSAduinoBridge_diego.ino文件中定义一个预编译符号,这个预编译符号可以启动或者关闭4驱的代码,这样方便我们开关4驱的功能

#define L298P_4WD

在diego 1 4wd中实现了使用arduino uno读取4个马达的AB项编码输出,一般的网上说明中arduino uno只有两个外部中断,好像只能读取一个马达的AB项编码输出,但事实上arudino中所有IO引脚都可以作为中断使用,通过操作中断寄存器的方式。

首先我们在encoder_driver.h文件中定义编码器连接的Arduino uno引脚:

#ifdef ARDUINO_ENC_COUNTER
  //below can be changed, but should be PORTD pins; 
  //otherwise additional changes in the code are required
  #define LEFT_ENC_PIN_A PD2  //pin 2
  #define LEFT_ENC_PIN_B PD3  //pin 3
  
  //below can be changed, but should be PORTC pins
  #define RIGHT_ENC_PIN_A PC2  //pin A2
  #define RIGHT_ENC_PIN_B PC3   //pin A3

#ifdef L298P_4WD
  #define LEFT_H_ENC_PIN_A PD4  //pin 4
  #define LEFT_H_ENC_PIN_B PD5  //pin 5
  
  //below can be changed, but should be PORTC pins
  #define RIGHT_H_ENC_PIN_A PC0  //pin A0
  #define RIGHT_H_ENC_PIN_B PC1   //pin A1
#endif  
#endif

从此文件中可以看到在原来2驱的基础上增加了4WD模式下的编码器连接引脚定义,其中左后方的电机AB项连接数字IO的D4和D5,而右后方电机AB项连接模拟IO的A0和A1,这样4个马达的编码器数据读取我们就用到8个IO口,加上PWM控制,转动方向的控制,在Diego 1# 4WD版本中,底盘控制一共用了16个IO,最后只剩D0,D1作为串口和上位机通讯,和A4,A5作为I2C的接口与I2C接口模块通讯,可以说Arduino UNO做到了充分利用。

在encoder_driver.ino文件中增加对4WD新增引脚的中断处理。

arduino uno中一共有3个引脚中断函数分别是

  • ISR (PCINT0_vect)对应 D8 to D13
  • ISR (PCINT1_vect) 对应 A0 to A5
  • ISR (PCINT2_vect) 对应 D0 to D7

diego1# 4wd中只需要用到两个中断处理函数 ISR(PCINT2_vect)和ISR(PCINT1_vect),代码如下:

  ISR (PCINT2_vect){
     static uint8_t enc_last=0;
#ifdef L298P_4WD        
     static uint8_t enc_last_h=0;
#endif          
     enc_last <<=2; //shift previous state two places
     enc_last |= (PIND & (3 << 2)) >> 2; //read the current state into lowest 2 bits

#ifdef L298P_4WD
     enc_last_h<<=2;
     enc_last_h |=(PIND & (3 << 4))>>4;
#endif 
  
     left_enc_pos += ENC_STATES[(enc_last & 0x0f)];
#ifdef L298P_4WD   
     left_h_enc_pos +=ENC_STATES[(enc_last_h & 0x0f)];
#endif    
  }
  
  /* Interrupt routine for RIGHT encoder, taking care of actual counting */
  ISR (PCINT1_vect){
     static uint8_t enc_last=0;
#ifdef L298P_4WD        
     //uint8_t pinct=PINC;
     static uint8_t enc_last_h=0;
#endif   	
     enc_last <<=2; //shift previous state two places
     enc_last |= (PINC & (3 << 2)) >> 2; //read the current state into lowest 2 bits

#ifdef L298P_4WD
     enc_last_h<<=2;
     enc_last_h |=(PINC & 3);
#endif  
     right_enc_pos += ENC_STATES[(enc_last & 0x0f)];
#ifdef L298P_4WD   
     right_h_enc_pos +=ENC_STATES[(enc_last_h & 0x0f)];
#endif
  }

这段代码中主要是针对中断寄存器PINC和PIND的操作,每个IO pin都对应PINC或者PIND的一位,IO有中断产生时,对应的位就会被置位,我们只需要读取相应为即可,如(PIND & (3 << 4))>>4读取的就是PIND的第4,5位,也就是D4,D5的数据。

Lesson 21:moveit-diego1# plus 的moveit驱动

在上节课中我们已经通过moveit assistant完成了针对diego 1# plus的配置,但如果想让机械臂动起来,我们还需要对编写驱动代码:代码原理如下:

首先,moveit把计算的结果通过Ros action的方式发送给driver,driver调用Ros_arduino_bridge的servor_write server发送各个关节舵机的控制指令给Arduino uno控制板
其次,同时Driver也要通过调用Ros_arduino_bridge的servo_read服务读取各个关节的舵机状态,通过joint_state消息的方式发送给moveit,供moveit进行路径规划计算。

接下来我们一步一步的来完成驱动

1.创建控制器配置文件diego_controllers.yaml

在moveit assistant产生的配置文件目录的config子目录下,创建Diego 1# plus的配置文件diego1_controllers.yaml,此文件告诉moveit将与哪个的action通讯

diego1_controllers.yaml代码如下:

controller_list:
  - name: arm_controller
    action_ns: follow_joint_trajectory
    type: FollowJointTrajectory
    default: true
    joints:
      - arm_base_to_arm_round_joint_stevo0
      - shoulder_2_to_arm_joint_stevo1
      - big_arm_round_to_joint_stevo2
      - arm_joint_stevo2_to_arm_joint_stevo3
      - wrist_to_arm_joint_stevo4
      - arm_joint_stevo4_to_arm_joint_stevo5
  - name: gripper_controller
    action_ns: gripper_action
    type: GripperCommand
    default: true
    joints:
      - hand_to_grip_joint_stevo6
      - hand_to_grip_joint_stevo7

这里分别为机械臂和机械爪定义了两个控制器arm_controller和gripper_controller,告诉moveit机械臂通讯的topic是arm_controller/follow_joint_trajectory,机械爪通讯的topic是gripper_controller/gripper_action

2.机械臂控制的FollowController

我们扩展了ros_arduino_bridge的功能来实现FollowController,在ros_arduino_python目录下创建joints.py和diego1_follow_controller.py

2.1 joint.py

封装了关节舵机控制的代码,主要功能函数:

  • __init__类初始化代码
  • setCurrentPosition(self):通过调用/arduino/servo_write服务发送舵机的控制命令,把舵机的位置设置为self.position
  • setCurrentPosition(self, position):
    通过调用/arduino/servo_write服务发送舵机的控制命令,把舵机的位置设置为moveit给出的舵机位置
  • mapToServoPosition(self,position):将moveit给出的舵机位置,转换为机械臂舵机实际对应的位置,此功能需要标定后,才能准确控制机械臂的位置。
#!/usr/bin/env python

# Copyright (c) 2017-2018 diego. 
# All right reserved.
#

## @file joints.py the jonit clase support functions for joints.

## @brief Joints hold current values.
import roslib
import rospy
from ros_arduino_msgs.srv import *
from math import pi as PI, degrees, radians
class Joint:

    ## @brief Constructs a Joint instance.
    ##
    ## @param servoNum The servo id for this joint.
    ## 
    ## @param name The joint name.
    ## 
    ## @param name The servo control range.
    def __init__(self, name, servoNum, max, min, servo_max, servo_min, inverse):
        self.name = name
        self.servoNum=servoNum
        self.max=max
        self.min=min
        self.servo_max=servo_max
        self.servo_min=servo_min
        self.inverse=inverse

        self.position = 0.0
        self.velocity = 0.0

    ## @brief Set the current position.
    def setCurrentPosition(self):
        rospy.wait_for_service('/arduino/servo_write')
	try:
	        servo_write=rospy.ServiceProxy('/arduino/servo_write',ServoWrite)
	        servo_write(self.servoNum,self.position)
        except rospy.ServiceException, e:
            print "Service call failed: %s"%e   

    ## @brief Set the current position.
    ##
    ## @param position The current position. 
    def setCurrentPosition(self, position):
        rospy.wait_for_service('/arduino/servo_write')
	try:
		#if self.servoNum==2:
        	servo_write=rospy.ServiceProxy('/arduino/servo_write',ServoWrite)
        	self.mapToServoPosition(position)	        	
       		servo_write(self.servoNum,radians(self.position))

        except rospy.ServiceException, e:
            print "Service call failed: %s"%e
    
    ## @brief map to  Servo Position.
    ##
    ## @param position The current position.        
    def mapToServoPosition(self,position):
	    per=(position-self.min)/(self.max-self.min)
	    if not self.inverse:	    	
	    	self.position=self.servo_min+per*(self.servo_max-self.servo_min)
	    else:
	    	self.position=self.servo_max-per*(self.servo_max-self.servo_min)


2.2diego1_follow_controller.py

此类diego机械臂控制器类,接收moveit的控制命令,转换为关节舵机的实际控制指令,驱动机械臂的运动。

#!/usr/bin/env python

# Copyright (c) 2017-2018 williamar. 
# All right reserved.

import rospy, actionlib

from control_msgs.msg import FollowJointTrajectoryAction
from trajectory_msgs.msg import JointTrajectory
from diagnostic_msgs.msg import *
from math import pi as PI, degrees, radians
from joints import Joint

class FollowController:

    def __init__(self, name):
        self.name = name
        
        rospy.init_node(self.name)
        

        # rates
        self.rate = 10.0
        
        # Arm jonits
        self.arm_base_to_arm_round_joint_stevo0=Joint('arm_base_to_arm_round_joint_stevo0',0,1.5797,-1.5707,130,0,False)
        self.shoulder_2_to_arm_joint_stevo1=Joint('shoulder_2_to_arm_joint_stevo1',1,1.5707,-0.1899,115,45,False)
        self.big_arm_round_to_joint_stevo2=Joint('big_arm_round_to_joint_stevo2',2,2.5891,1,100,20,False)
        self.arm_joint_stevo2_to_arm_joint_stevo3=Joint('arm_joint_stevo2_to_arm_joint_stevo3',3,1.5707,-1.5707,130,0,False)
        self.wrist_to_arm_joint_stevo4=Joint('wrist_to_arm_joint_stevo4',4,1.5707,-1.5707,130,0,False)
        self.arm_joint_stevo4_to_arm_joint_stevo5=Joint('arm_joint_stevo4_to_arm_joint_stevo5',5,1.5707,-1.5707,130,0,True)
        
        
        
        self.joints=[self.arm_base_to_arm_round_joint_stevo0,
        self.shoulder_2_to_arm_joint_stevo1,
        self.big_arm_round_to_joint_stevo2,
        self.arm_joint_stevo2_to_arm_joint_stevo3,
        self.wrist_to_arm_joint_stevo4,
        self.arm_joint_stevo4_to_arm_joint_stevo5]
        
        # action server
        self.server = actionlib.SimpleActionServer('arm_controller/follow_joint_trajectory', FollowJointTrajectoryAction, execute_cb=self.actionCb, auto_start=False)
        self.server.start()
        rospy.spin()
        rospy.loginfo("Started FollowController")


    def actionCb(self, goal):
	print("****************actionCb*********************")    
        rospy.loginfo(self.name + ": Action goal recieved.")
        traj = goal.trajectory

        if set(self.joints) != set(traj.joint_names):
            for j in self.joints:
                if j.name not in traj.joint_names:
                    msg = "Trajectory joint names does not match action controlled joints." + str(traj.joint_names)
                    rospy.logerr(msg)
                    self.server.set_aborted(text=msg)
                    return
            rospy.logwarn("Extra joints in trajectory")

        if not traj.points:
            msg = "Trajectory empy."
            rospy.logerr(msg)
            self.server.set_aborted(text=msg)
            return

        try:
            indexes = [traj.joint_names.index(joint.name) for joint in self.joints]
        except ValueError as val:
            msg = "Trajectory invalid."
            rospy.logerr(msg)
            self.server.set_aborted(text=msg)
            return

        if self.executeTrajectory(traj):   
            self.server.set_succeeded()
        else:
            self.server.set_aborted(text="Execution failed.")

        rospy.loginfo(self.name + ": Done.")     

    def executeTrajectory(self, traj):
        rospy.loginfo("Executing trajectory")
        rospy.logdebug(traj)
        # carry out trajectory
        try:
            indexes = [traj.joint_names.index(joint.name) for joint in self.joints]
        except ValueError as val:
            rospy.logerr("Invalid joint in trajectory.")
            return False

        # get starting timestamp, MoveIt uses 0, need to fill in
        start = traj.header.stamp
        if start.secs == 0 and start.nsecs == 0:
            start = rospy.Time.now()

        r = rospy.Rate(self.rate)
	for point in traj.points:            
            desired = [ point.positions[k] for k in indexes ]
            for i in indexes:
                #self.joints[i].position=desired[i]
                self.joints[i].setCurrentPosition(desired[i])

            while rospy.Time.now() + rospy.Duration(0.01) < start:
                rospy.sleep(0.01)
        return True
    

if __name__=='__main__':
    try:
        rospy.loginfo("start followController...")
        FollowController('follow_Controller')
    except rospy.ROSInterruptException:
        rospy.loginfo('Failed to start followController...')
        


此类是驱动的核心代码,接下来我们详细解释其功能逻辑

    def __init__(self, name):
        self.name = name
        
        rospy.init_node(self.name)
        

        # rates
        self.rate = 10.0

初始化化ros note,设置节点名称,刷新频率为10hz

        # Arm jonits
        self.arm_base_to_arm_round_joint_stevo0=Joint('arm_base_to_arm_round_joint_stevo0',0,1.5797,-1.5707,130,0,False)
        self.shoulder_2_to_arm_joint_stevo1=Joint('shoulder_2_to_arm_joint_stevo1',1,1.5707,-0.1899,115,45,False)
        self.big_arm_round_to_joint_stevo2=Joint('big_arm_round_to_joint_stevo2',2,2.5891,1,100,20,False)
        self.arm_joint_stevo2_to_arm_joint_stevo3=Joint('arm_joint_stevo2_to_arm_joint_stevo3',3,1.5707,-1.5707,130,0,False)
        self.wrist_to_arm_joint_stevo4=Joint('wrist_to_arm_joint_stevo4',4,1.5707,-1.5707,130,0,False)
        self.arm_joint_stevo4_to_arm_joint_stevo5=Joint('arm_joint_stevo4_to_arm_joint_stevo5',5,1.5707,-1.5707,130,0,True)
        
        
        
        self.joints=[self.arm_base_to_arm_round_joint_stevo0,
        self.shoulder_2_to_arm_joint_stevo1,
        self.big_arm_round_to_joint_stevo2,
        self.arm_joint_stevo2_to_arm_joint_stevo3,
        self.wrist_to_arm_joint_stevo4,
        self.arm_joint_stevo4_to_arm_joint_stevo5]

初始化机械臂的关节,并把关节放入joints列表中,方便后续操作

        # action server
        self.server = actionlib.SimpleActionServer('arm_controller/follow_joint_trajectory', FollowJointTrajectoryAction, execute_cb=self.actionCb, auto_start=False)
        self.server.start()
        rospy.spin()
        rospy.loginfo("Started FollowController")

定义action server并启动。
指定通讯的topic为arm_controller/follow_joint_trajectory,类型为FollowJointTrajectoryAction,收到消息后的回调函数为self.actionCb,
self.actionCb的其实就是follow_controller对JointTrajectory msg的处理,所以这里先介绍一下JointTrajectory msg,只要理解了JointTrajectory msg,有助与后续代码的理解,执行如下命令就可以显示了JointTrajectory msg的结构。

$ rosmsg show JointTrajectory

可以看到消息的结构体中包含了三部分

  •  header  这是Ros的标准消息头这里就不多介绍了
  •  joint_names 这是所有关节名称的数组
  • JointTrajectoryPoint 这部分是驱动的关键,这个数组记录了机械臂从一种姿势到另外一种姿势所经过的路径点,moveit所产生的姿势路径是通过这些point点描述出来的,也就是我们驱动中要控制每个关节的舵机都按照这些point点进行运动,每个point又是由一个结构体构成:
  • positions这是一个float64的数组,记录每个point的时候舵机应该到达的角度,这里是弧度为单位的,比如说是6自由度的那每个Point的这个positions字段中应该包含六个数值[1.57,0,2,0.2,3,0.12],也就是我们舵机控制范围是180度,那这里面的取值范围就是0~π
  • velocities这个数组记录了每个关节运动的速度
  • accelerations这个数组记录每个关节运动的加速度
  • effort这个参数可以不用
  • time_from_start这个参数是指定从头部的timestamp开始算起多长时间要达到这个点的位置

理解了JointTrajectory msg的结构,我们很容易理解函数self.actionCb就是对消息做逻辑上的异常处理,判断msg中joint是否在joints列表中,消息是否为空,而真正消息体的处理是在executeTrajectory(self, traj),下面我们来看这个函数的处理逻辑

    def executeTrajectory(self, traj):
        rospy.loginfo("Executing trajectory")
        rospy.logdebug(traj)
        # carry out trajectory
        try:
            indexes = [traj.joint_names.index(joint.name) for joint in self.joints]
        except ValueError as val:
            rospy.logerr("Invalid joint in trajectory.")
            return False

将JointTrajectory msg中的Joint按照self.joints的顺序索引,并将索引放入indexes数组中。

    # get starting timestamp, MoveIt uses 0, need to fill in
        start = traj.header.stamp
        if start.secs == 0 and start.nsecs == 0:
            start = rospy.Time.now()

获取当前msg的时间戳

r = rospy.Rate(self.rate)

设定刷新频率

	for point in traj.points:            
            desired = [ point.positions[k] for k in indexes ]
            for i in indexes:
                #self.joints[i].position=desired[i]
                self.joints[i].setCurrentPosition(desired[i])

            while rospy.Time.now() + rospy.Duration(0.01) < start:
                rospy.sleep(0.01)

循环读取msg中的位置点,位置点中的舵机位置值读取出来,调用Joint类的setCurrentPosition方法发送舵机的控制指令

3. launch启动文件

moveit assistant会生成一个launch文件夹,其中包含了主要的launch文件,我们只需要做些接单的修改,即可进行测试。

3.1修改diego1_moveit_controller_manager.launch.xml

此文件是moveit assistant自动生成的,但其中内容是空的,我增加如下内容,告诉moveit,启动Controller的配置文件位置

<launch>
     <!-- Set the param that trajectory_execution_manager needs to find the controller plugin -->
     <arg name="moveit_controller_manager" default="moveit_simple_controller_manager/MoveItSimpleControllerManager" />
     <param name="moveit_controller_manager" value="$(arg moveit_controller_manager)"/>
     <!-- load controller_list -->
     <rosparam file="$(find diego_moveit_config)/config/diego1_controllers.yaml"/>
</launch>

3.2创建diego_demo.launch文件

在moveit的launch文件夹下创建diego_demo.launch文件,并添加如下内容

<launch>

     <master auto="start"/>

     <node name="arduino" pkg="ros_arduino_python" type="arduino_node.py" output="screen">
         <rosparam file="$(find ros_arduino_python)/config/my_arduino_params.yaml" command="load" />
     </node>
     <!-- By default, we are not in debug mode -->
     <arg name="debug" default="false" />

     <!-- Load the URDF, SRDF and other .yaml configuration files on the param server -->
     <include file="$(find diego_moveit_config)/launch/planning_context.launch">
         <arg name="load_robot_description" value="true"/>
     </include>

     <node name="arduino_follow_controller" pkg="ros_arduino_python" type="diego1_follow_controller.py" output="screen">
     </node>

     <!-- If needed, broadcast static tf for robot root -->
     <node pkg="tf" type="static_transform_publisher" name="virtual_joint_broadcaster_0" args="0 0 0 0 0 0 world_frame base_link 100" />

     <!-- We do not have a robot connected, so publish fake joint states -->
     <node name="joint_state_publisher" pkg="joint_state_publisher" type="joint_state_publisher">
        <param name="/use_gui" value="false"/>
        <rosparam param="/source_list">[/move_group/fake_controller_joint_states]</rosparam>
     </node>

     <!-- Run the main MoveIt executable without trajectory execution (we do not have controllers configured by default) -->
     <include file="$(find diego_moveit_config)/launch/move_group.launch">
     </include>
</launch>

这里主要是作为一个测试使用,所以joint_states使用了假数据,在调用joint_state_publisher时我们把source list设置为/move_group/fake_controller_joint_states

现在我们可以执行如下代码来启动moveit

roslaunch diego_moveit_config diego_demo.launch

这时候我们可以通过控制台输入moveit命令来控制机械臂,我们也可以用代码来调用moveit接口来控制机械臂,下面我们来写一段测试代码来

4.moveit控制测试代码

在ros_arduino_python目录下创建diego_moveit_test.py,代码如下

#!/usr/bin/env python

import rospy, sys
import moveit_commander
from control_msgs.msg import GripperCommand

class MoveItDemo:
    def __init__(self):
        # Initialize the move_group API
        moveit_commander.roscpp_initialize(sys.argv)

        # Initialize the ROS node
        rospy.init_node('moveit_demo', anonymous=True)
        
 
        # Connect to the arm move group
        arm = moveit_commander.MoveGroupCommander('arm')
                       
        # Get the name of the end-effector link
        end_effector_link = arm.get_end_effector_link()
        
        # Display the name of the end_effector link
        rospy.loginfo("The end effector link is: " + str(end_effector_link))
        
        # Set a small tolerance on joint angles
        arm.set_goal_joint_tolerance(0.001)
        
        # Start the arm target in "resting" pose stored in the SRDF file
        arm.set_named_target('arm_default_pose')
        
        # Plan a trajectory to the goal configuration
        traj = arm.plan()
         
        # Execute the planned trajectory
        arm.execute(traj)
        
        # Pause for a moment
        rospy.sleep(1)         
        
        # Cleanly shut down MoveIt
        moveit_commander.roscpp_shutdown()
        
        # Exit the script
        moveit_commander.os._exit(0)

if __name__ == "__main__":
    try:
        MoveItDemo()
    except rospy.ROSInterruptException:
        pass

在测试代码中,我们连接到group arm,并命令机械臂运行到姿势 arm_default_pose,我们运行如下命令启动测试代码

    rosrun ros_arduino_python diego_moveit_test.py

启动后可以看到机械臂马上响应到arm_default_pose的位置。
 

Lesson 18:机器视觉-人脸跟踪

基于前面两节人脸检测,特征值获取及跟踪,我们进一步实现人脸的跟踪,我们可以让机器人跟踪人脸,暨当人脸移动时控制机器人转动,使得人脸始终在图像窗体的中间位置,基本流程如下

 

 

 

 

  • 人脸检测,lesson 16中已经介绍过了,这里直接引用就可以。
  • 特征获取,lesson 17中也已经实现,但要达到比较好的效果,需要对特征点做一些补偿及噪声点的剔除
  • 人脸跟踪,需要根据ROI的位置,来判断机器人需要怎样移动才能达到跟踪的效果

1. 人脸检测及特征获取

源文件face_tracker2.py请见github,代码已经在opencv3环境完成测试

FaceTracker继承FaceDetector, LKTracker两个类,并做了如下的主要修改和扩展

  • 重写process_image函数
  • add_keypoints函数用于发现新的特征点,其调用goodFeaturesToTrack方法,并判断与当前特征点cluster的距离,以保证添加有效的新特征点
  • drop_keypoints函数采用聚类算法,将无效的特征点剔除

1.1 源代码

#!/usr/bin/env python

import roslib
import rospy
import cv2
import numpy as np
from face_detector import FaceDetector
from lk_tracker import LKTracker

class FaceTracker(FaceDetector, LKTracker):
    def __init__(self, node_name):
        super(FaceTracker, self).__init__(node_name)
        
        self.n_faces = rospy.get_param("~n_faces", 1)
        self.show_text = rospy.get_param("~show_text", True)
        self.show_add_drop = rospy.get_param("~show_add_drop", False)
        self.feature_size = rospy.get_param("~feature_size", 1)
        self.use_depth_for_tracking = rospy.get_param("~use_depth_for_tracking", False)
        self.min_keypoints = rospy.get_param("~min_keypoints", 20)
        self.abs_min_keypoints = rospy.get_param("~abs_min_keypoints", 6)
        self.std_err_xy = rospy.get_param("~std_err_xy", 2.5) 
        self.pct_err_z = rospy.get_param("~pct_err_z", 0.42) 
        self.max_mse = rospy.get_param("~max_mse", 10000)
        self.add_keypoint_distance = rospy.get_param("~add_keypoint_distance", 10)
        self.add_keypoints_interval = rospy.get_param("~add_keypoints_interval", 1)
        self.drop_keypoints_interval = rospy.get_param("~drop_keypoints_interval", 1)
        self.expand_roi_init = rospy.get_param("~expand_roi", 1.02)
        self.expand_roi = self.expand_roi_init
        self.face_tracking = True

        self.frame_index = 0
        self.add_index = 0
        self.drop_index = 0
        self.keypoints = list()

        self.detect_box = None
        self.track_box = None
        
        self.grey = None
        self.prev_grey = None
        
    def process_image(self, cv_image):
        try:
            # Create a greyscale version of the image
            self.grey = cv2.cvtColor(cv_image, cv2.COLOR_BGR2GRAY)
            
            # Equalize the grey histogram to minimize lighting effects
            self.grey = cv2.equalizeHist(self.grey)
            
            # Step 1: Detect the face if we haven't already
            if self.detect_box is None:
                self.keypoints = list()
                self.track_box = None
                self.detect_box = self.detect_face(self.grey)
            else:
                # Step 2: If we aren't yet tracking keypoints, get them now
                if not self.track_box or not self.is_rect_nonzero(self.track_box):
                    self.track_box = self.detect_box
                    self.keypoints = self.get_keypoints(self.grey, self.track_box)
    
                # Store a copy of the current grey image used for LK tracking                   
                if self.prev_grey is None:
                    self.prev_grey = self.grey           
                  
                # Step 3: If we have keypoints, track them using optical flow
                self.track_box = self.track_keypoints(self.grey, self.prev_grey)
              
                # Step 4: Drop keypoints that are too far from the main cluster
                if self.frame_index % self.drop_keypoints_interval == 0 and len(self.keypoints) > 0:
                    ((cog_x, cog_y, cog_z), mse_xy, mse_z, score) = self.drop_keypoints(self.abs_min_keypoints, self.std_err_xy, self.max_mse)
                    
                    if score == -1:
                        self.detect_box = None
                        self.track_box = None
                        return cv_image
                  
                # Step 5: Add keypoints if the number is getting too low 
                if self.frame_index % self.add_keypoints_interval == 0 and len(self.keypoints) < self.min_keypoints:
                    self.expand_roi = self.expand_roi_init * self.expand_roi
                    self.add_keypoints(self.track_box)
                else:
                    self.frame_index += 1
                    self.expand_roi = self.expand_roi_init
            
            # Store a copy of the current grey image used for LK tracking            
            self.prev_grey = self.grey
              
            # Process any special keyboard commands for this module
            self.prev_grey = self.grey
              
            # Process any special keyboard commands for this module
            if 32 <= self.keystroke and self.keystroke < 128:
                cc = chr(self.keystroke).lower()
                if cc == 'c':
                    self.keypoints = []
                    self.track_box = None
                    self.detect_box = None
                elif cc == 'd':
                    self.show_add_drop = not self.show_add_drop
                    
        except AttributeError:
            pass
                    
        return cv_image

    def add_keypoints(self, track_box):
        # Look for any new keypoints around the current keypoints
        
        # Begin with a mask of all black pixels
        mask = np.zeros_like(self.grey)
        
        # Get the coordinates and dimensions of the current track box
        try:
            ((x,y), (w,h), a) = track_box
        except:
            try:
                x,y,w,h = track_box
            except:
                rospy.loginfo("Track box has shrunk to zero...")
                return
        
        x = int(x)
        y = int(y)
        
        # Expand the track box to look for new keypoints
        w_new = int(self.expand_roi * w)
        h_new = int(self.expand_roi * h)
        
        pt1 = (x - int(w_new / 2), y - int(h_new / 2))
        pt2 = (x + int(w_new / 2), y + int(h_new / 2))
        
        mask_box = ((x, y), (w_new, h_new), a)

        # Display the expanded ROI with a yellow rectangle
        if self.show_add_drop:
            cv2.rectangle(self.marker_image, pt1, pt2, (255, 255, 0))
                        
        # Create a filled white ellipse within the track_box to define the ROI
        cv2.ellipse(mask, mask_box, (255,255, 255), cv2.FILLED)
        
        if self.keypoints is not None:
            # Mask the current keypoints
            for x, y in [np.int32(p) for p in self.keypoints]:
                cv2.circle(mask, (x, y), 5, 0, -1)
         
        new_keypoints = cv2.goodFeaturesToTrack(self.grey, mask = mask, **self.gf_params)

        # Append new keypoints to the current list if they are not
        # too far from the current cluster      
        if new_keypoints is not None:
            for x, y in np.float32(new_keypoints).reshape(-1, 2):
                distance = self.distance_to_cluster((x,y), self.keypoints)
                if distance > self.add_keypoint_distance:
                    self.keypoints.append((x,y))
                    # Briefly display a blue disc where the new point is added
                    if self.show_add_drop:
                        cv2.circle(self.marker_image, (x, y), 3, (255, 255, 0, 0), cv2.FILLED, 2, 0)
                                    
            # Remove duplicate keypoints
            self.keypoints = list(set(self.keypoints))
        
    def distance_to_cluster(self, test_point, cluster):
        min_distance = 10000
        for point in cluster:
            if point == test_point:
                continue
            # Use L1 distance since it is faster than L2
            distance = abs(test_point[0] - point[0])  + abs(test_point[1] - point[1])
            if distance < min_distance:
                min_distance = distance
        return min_distance

    def drop_keypoints(self, min_keypoints, outlier_threshold, mse_threshold):
        sum_x = 0
        sum_y = 0
        sum_z = 0
        sse = 0
        keypoints_xy = self.keypoints
        keypoints_z = self.keypoints
        n_xy = len(self.keypoints)
        n_z = n_xy
        
        if self.use_depth_for_tracking:
            if self.depth_image is None:
                return ((0, 0, 0), 0, 0, -1)
        
        # If there are no keypoints left to track, start over
        if n_xy == 0:
            return ((0, 0, 0), 0, 0, -1)
        
        # Compute the COG (center of gravity) of the cluster
        for point in self.keypoints:
            sum_x = sum_x + point[0]
            sum_y = sum_y + point[1]
        
        mean_x = sum_x / n_xy
        mean_y = sum_y / n_xy
        
        if self.use_depth_for_tracking:
            for point in self.keypoints:
                try:
                    z = cv2.get2D(self.depth_image, min(self.frame_height - 1, int(point[1])), min(self.frame_width - 1, int(point[0])))
                except:
                    continue
                z = z[0]
                # Depth values can be NaN which should be ignored
                if isnan(z):
                    continue
                else:
                    sum_z = sum_z + z
                    
            mean_z = sum_z / n_z
            
        else:
            mean_z = -1
        
        # Compute the x-y MSE (mean squared error) of the cluster in the camera plane
        for point in self.keypoints:
            sse = sse + (point[0] - mean_x) * (point[0] - mean_x) + (point[1] - mean_y) * (point[1] - mean_y)
            #sse = sse + abs((point[0] - mean_x)) + abs((point[1] - mean_y))
        
        # Get the average over the number of feature points
        mse_xy = sse / n_xy

        # The MSE must be > 0 for any sensible feature cluster
        if mse_xy == 0 or mse_xy > mse_threshold:
            return ((0, 0, 0), 0, 0, -1)
        
        # Throw away the outliers based on the x-y variance
        max_err = 0
        for point in self.keypoints:
            std_err = ((point[0] - mean_x) * (point[0] - mean_x) + (point[1] - mean_y) * (point[1] - mean_y)) / mse_xy
            if std_err > max_err:
                max_err = std_err
            if std_err > outlier_threshold:
                keypoints_xy.remove(point)
                if self.show_add_drop:
                    # Briefly mark the removed points in red
                    cv2.circle(self.marker_image, (point[0], point[1]), 3, (0, 0, 255), cv2.FILLED, 2, 0)   
                try:
                    keypoints_z.remove(point)
                    n_z = n_z - 1
                except:
                    pass
                
                n_xy = n_xy - 1
                                
        # Now do the same for depth
        if self.use_depth_for_tracking:
            sse = 0
            for point in keypoints_z:
                try:
                    z = cv2.get2D(self.depth_image, min(self.frame_height - 1, int(point[1])), min(self.frame_width - 1, int(point[0])))
                    z = z[0]
                    sse = sse + (z - mean_z) * (z - mean_z)
                except:
                    n_z = n_z - 1
            
            if n_z != 0:
                mse_z = sse / n_z
            else:
                mse_z = 0
            
            # Throw away the outliers based on depth using percent error 
            # rather than standard error since depth values can jump
            # dramatically at object boundaries
            for point in keypoints_z:
                try:
                    z = cv2.get2D(self.depth_image, min(self.frame_height - 1, int(point[1])), min(self.frame_width - 1, int(point[0])))
                    z = z[0]
                except:
                    continue
                try:
                    pct_err = abs(z - mean_z) / mean_z
                    if pct_err > self.pct_err_z:
                        keypoints_xy.remove(point)
                        if self.show_add_drop:
                            # Briefly mark the removed points in red
                            cv2.circle(self.marker_image, (point[0], point[1]), 2, (0, 0, 255), cv2.FILLED)  
                except:
                    pass
        else:
            mse_z = -1
        
        self.keypoints = keypoints_xy
               
        # Consider a cluster bad if we have fewer than min_keypoints left
        if len(self.keypoints) < min_keypoints:
            score = -1
        else:
            score = 1

        return ((mean_x, mean_y, mean_z), mse_xy, mse_z, score)
    
if __name__ == '__main__':
    try:
        node_name = "face_tracker"
        FaceTracker(node_name)
        rospy.spin()
    except KeyboardInterrupt:
        print "Shutting down face tracker node."
        cv2.destroyAllWindows()

1.2 launch文件

<launch>
   <node pkg="diego_vision" name="face_tracker2" type="face_tracker2.py" output="screen">

   <remap from="input_rgb_image" to="/camera/rgb/image_color" />
   <remap from="input_depth_image" to="/camera/depth/image" />
 
   <rosparam>
       use_depth_for_tracking: True
       min_keypoints: 20
       abs_min_keypoints: 6
       add_keypoint_distance: 10
       std_err_xy: 2.5
       pct_err_z: 0.42
       max_mse: 10000
       add_keypoints_interval: 1
       drop_keypoints_interval: 1
       show_text: True
       show_features: True
       show_add_drop: False
       feature_size: 1
       expand_roi: 1.02
       gf_maxCorners: 200
       gf_qualityLevel: 0.02
       gf_minDistance: 7
       gf_blockSize: 10
       gf_useHarrisDetector: False
       gf_k: 0.04
       haar_scaleFactor: 1.3
       haar_minNeighbors: 3
       haar_minSize: 30
       haar_maxSize: 150
   </rosparam>

   <param name="cascade_1" value="$(find diego_vision)/data/haar_detectors/haarcascade_frontalface_alt2.xml" />
   <param name="cascade_2" value="$(find diego_vision)/data/haar_detectors/haarcascade_frontalface_alt.xml" />
   <param name="cascade_3" value="$(find diego_vision)/data/haar_detectors/haarcascade_profileface.xml" />

</node>
</launch>

在Diego1 plus中使用深度相机,故把use_depth_for_tracking参数设置为True,实践表明使用深度相机会有更好的效果,这时我们分别在两个terminal中启动openni节点,和face_tracker2.py节点,会出现视频窗口,如果有人脸出现在面前就会被捕捉到,并被标识出来,而且效果也不错。没有想到是Diego1 plus的配置下,人脸识别任然会感觉到有滞后的现象。

roslaunch diego_vision openni_node.launch

roslaucn diego_vision face_tracker2.launch

2.人脸跟踪

源文件object_tracker2.py请见github,代码已经在opencv3环境完成测试

ObjectTracker类主要功能如下:

  • 订阅ROI信息,此消息来由FaceTracker类发布,表示捕捉到的人脸的位置,ObjectTracker订阅此消息来判断人脸在画面中的位置
  • 订阅/camera/depth/image消息,判断人脸距离摄像头的位置,以实现跟踪
  • 发布Twist消息,控制机器人的移动

2.1 源代码

#!/usr/bin/env python

import roslib
import rospy
from sensor_msgs.msg import Image, RegionOfInterest, CameraInfo
from geometry_msgs.msg import Twist
from math import copysign, isnan
from cv_bridge import CvBridge, CvBridgeError
import numpy as np

class ObjectTracker():
    def __init__(self):
        rospy.init_node("object_tracker")
                        
        # Set the shutdown function (stop the robot)
        rospy.on_shutdown(self.shutdown)
        
        # How often should we update the robot's motion?
        self.rate = rospy.get_param("~rate", 10)
        r = rospy.Rate(self.rate) 
        
        # The maximum rotation speed in radians per second
        self.max_rotation_speed = rospy.get_param("~max_rotation_speed", 2.0)
        
        # The minimum rotation speed in radians per second
        self.min_rotation_speed = rospy.get_param("~min_rotation_speed", 0.5)
        
        # The x threshold (% of image width) indicates how far off-center
        # the ROI needs to be in the x-direction before we react
        self.x_threshold = rospy.get_param("~x_threshold", 0.1)
        
        # The maximum distance a target can be from the robot for us to track
        self.max_z = rospy.get_param("~max_z", 2.0)
        
        # Initialize the global ROI
        self.roi = RegionOfInterest()
        
        # The goal distance (in meters) to keep between the robot and the person
        self.goal_z = rospy.get_param("~goal_z", 0.6)
        
        # How far away from the goal distance (in meters) before the robot reacts
        self.z_threshold = rospy.get_param("~z_threshold", 0.05)
        
        # How far away from being centered (x displacement) on the person
        # before the robot reacts
        self.x_threshold = rospy.get_param("~x_threshold", 0.05)
        
        # How much do we weight the goal distance (z) when making a movement
        self.z_scale = rospy.get_param("~z_scale", 1.0)

        # How much do we weight x-displacement of the person when making a movement        
        self.x_scale = rospy.get_param("~x_scale", 2.0)
        
        # The max linear speed in meters per second
        self.max_linear_speed = rospy.get_param("~max_linear_speed", 0.3)
        
        # The minimum linear speed in meters per second
        self.min_linear_speed = rospy.get_param("~min_linear_speed", 0.1)

        # Publisher to control the robot's movement
        self.cmd_vel_pub = rospy.Publisher('cmd_vel', Twist)
        
        # Intialize the movement command
        self.move_cmd = Twist()
        
        # We will get the image width and height from the camera_info topic
        self.image_width = 0
        self.image_height = 0
        
        # We need cv_bridge to convert the ROS depth image to an OpenCV array
        self.cv_bridge = CvBridge()
        self.depth_array = None
        
        # Set flag to indicate when the ROI stops updating
        self.target_visible = False
        
        # Wait for the camera_info topic to become available
        rospy.loginfo("Waiting for camera_info topic...")
        rospy.wait_for_message('camera_info', CameraInfo)
        
        # Subscribe to the camera_info topic to get the image width and height
        rospy.Subscriber('camera_info', CameraInfo, self.get_camera_info)

        # Wait until we actually have the camera data
        while self.image_width == 0 or self.image_height == 0:
            rospy.sleep(1)
                    
        # Subscribe to the registered depth image
        rospy.Subscriber("depth_image", Image, self.convert_depth_image)
        
        # Wait for the depth image to become available
        rospy.wait_for_message('depth_image', Image)
        
        # Subscribe to the ROI topic and set the callback to update the robot's motion
        rospy.Subscriber('roi', RegionOfInterest, self.set_cmd_vel)
        
        # Wait until we have an ROI to follow
        rospy.loginfo("Waiting for an ROI to track...")
        rospy.wait_for_message('roi', RegionOfInterest)
        
        rospy.loginfo("ROI messages detected. Starting tracker...")
        
        # Begin the tracking loop
        while not rospy.is_shutdown():
            # If the target is not visible, stop the robot
            if not self.target_visible:
                self.move_cmd = Twist()
            else:
                # Reset the flag to False by default
                self.target_visible = False
                
            # Send the Twist command to the robot
            self.cmd_vel_pub.publish(self.move_cmd)
            
            # Sleep for 1/self.rate seconds
            r.sleep()

    def set_cmd_vel(self, msg):
        # If the ROI has a width or height of 0, we have lost the target
        if msg.width == 0 or msg.height == 0:
            return
        
        # If the ROI stops updating this next statement will not happen
        self.target_visible = True
        
        self.roi = msg
        
        # Compute the displacement of the ROI from the center of the image
        target_offset_x = msg.x_offset + msg.width / 2 - self.image_width / 2

        try:
            percent_offset_x = float(target_offset_x) / (float(self.image_width) / 2.0)
        except:
            percent_offset_x = 0
            
        # Intialize the movement command
        self.move_cmd = Twist()
        
        # Rotate the robot only if the displacement of the target exceeds the threshold
        if abs(percent_offset_x) > self.x_threshold:
            # Set the rotation speed proportional to the displacement of the target
            try:
                speed = percent_offset_x * self.x_scale
                self.move_cmd.angular.z = -copysign(max(self.min_rotation_speed,
                                            min(self.max_rotation_speed, abs(speed))), speed)
            except:
                pass
            
        # Now compute the depth component
        n_z = sum_z = mean_z = 0
         
        # Get the min/max x and y values from the ROI
        min_x = self.roi.x_offset
        max_x = min_x + self.roi.width
        min_y = self.roi.y_offset
        max_y = min_y + self.roi.height
        
        # Get the average depth value over the ROI
        for x in range(min_x, max_x):
            for y in range(min_y, max_y):
                try:
                    z = self.depth_array[y, x]
                except:
                    continue
                
                # Depth values can be NaN which should be ignored
                if isnan(z) or z > self.max_z:
                    continue
                else:
                    sum_z = sum_z + z
                    n_z += 1   
        try:
            mean_z = sum_z / n_z
            
            if mean_z < self.max_z and (abs(mean_z - self.goal_z) > self.z_threshold):
                speed = (mean_z - self.goal_z) * self.z_scale
                self.move_cmd.linear.x = copysign(min(self.max_linear_speed, max(self.min_linear_speed, abs(speed))), speed)
        except:
            pass
                    
    def convert_depth_image(self, ros_image):
        # Use cv_bridge() to convert the ROS image to OpenCV format
        try:
            # The depth image is a single-channel float32 image
            depth_image = self.cv_bridge.imgmsg_to_cv2(ros_image, "32FC1")
        except CvBridgeError, e:
            print e

        # Convert the depth image to a Numpy array
        self.depth_array = np.array(depth_image, dtype=np.float32)

    def get_camera_info(self, msg):
        self.image_width = msg.width
        self.image_height = msg.height

    def shutdown(self):
        rospy.loginfo("Stopping the robot...")
        self.cmd_vel_pub.publish(Twist())
        rospy.sleep(1)     

if __name__ == '__main__':
    try:
        ObjectTracker()
        rospy.spin()
    except rospy.ROSInterruptException:
        rospy.loginfo("Object tracking node terminated.")

2.2 launch文件

<launch>
<param name="/camera/driver/depth_registration" value="True" />

<node pkg="diego_vision" name="object_tracker" type="object_tracker2.py" output="screen">

<remap from="camera_info" to="/camera/depth/camera_info" />
<remap from="depth_image" to="/camera/depth/image" />

<rosparam>
rate: 10
max_z: 2.0
goal_z: 0.6
z_threshold: 0.05
x_threshold: 0.1
z_scale: 1.0
x_scale: 0.1
max_rotation_speed: 0.2
min_rotation_speed: 0.02
max_linear_speed: 0.2
min_linear_speed: 0.05
</rosparam>

</node>
</launch>

在已经启动face_tracker2.py节点的情况下,执行如下代码启动object_tracker2.py节点,机器人就会跟随检测到的人脸移动。

roslaunch diego_vision object_tracker2.launch

需要特别说明的是在launch文件中有关速度的参数不能设置的太大,太大会导致机器人不停转来转去来跟踪人脸,原因是因为图像处理有滞后,速度乘以滞后时间后,就会导致机器人移动的位置超出了预期的位置,不停的矫正。

Posts navigation

1 2
Scroll to top