import cv2 import os from datetime import datetime import math import numpy as np import sys import random from pathlib import Path import time from scipy.spatial.transform import Rotation as R from scipy.interpolate import CubicSpline import matplotlib.pyplot as plt from mpl_toolkits.mplot3d import Axes3D sys.path.append(str(Path(__file__).resolve().parent.parent)) sys.path.append('/home/jsfb/jsfb_ws/MassageRobot_aubo/Massage/MassageControl') from hardware.remote_cam import ToolCamera from hardware.aubo_C5 import AuboC5 np.set_printoptions(precision=8, suppress=True) # 设置寻找亚像素角点的参数，采用的停止准则是最大循环次数30和最大误差容限0.001 criteria = (cv2.TERM_CRITERIA_MAX_ITER | cv2.TERM_CRITERIA_EPS, 30, 0.001) # 标定计算类 # -- 连接AuboC5 # -- 连接 ToolCamera # -- 运动到初始位置 计算下一步运动位姿 # -- 开始手眼标定 # -- 输出结果 class Calibration: def __init__(self, board_size, square_size, file_folder): """ 初始化标定计算类。 board_size: 标定板的格子数，格式为 (width, height) square_size: 格子的大小，单位为米 """ self.board_size = board_size self.square_size = square_size self.objp = self._create_object_points() self.image_points = [] self.robot_poses = [] self.poselist = [] self.cam = ToolCamera(host='127.0.0.1') self.cam.start() self.file_address = file_folder rgb_image, depth_image, camera_intrinsics = self.cam.get_latest_frame() cv2.imshow("rgb1",rgb_image) cv2.waitKey(0) if rgb_image is None : print(f'===============相机连接失败请检查并重试==============') else: print(f'===============相机连接成功==============') self.size = rgb_image.shape[:2] self.intrinsics = camera_intrinsics self.arm = AuboC5() self.arm.init() self.obj_points = [] print(f'===============机械臂连接成功==============') def _create_object_points(self): """创建标定板的3D点""" objp = np.zeros((self.board_size[0] * self.board_size[1], 3), np.float32) objp[:, :2] = np.mgrid[0:self.board_size[0], 0:self.board_size[1]].T.reshape(-1, 2) * self.square_size return objp def collect_data(self): #移动到初始位置 # move = [(90) * (np.pi / 180), 4.23 * (np.pi / 180), 124.33 * (np.pi / 180), 61.56 * (np.pi / 180), -89.61 * (np.pi / 180), 0 * (np.pi / 180)] move = [(89.74) * (np.pi / 180), 11.21 * (np.pi / 180), 97.12 * (np.pi / 180), 29.06 * (np.pi / 180), -89.61 * (np.pi / 180), 0 * (np.pi / 180)] code = self.arm.move_joint(move,4,wait=True) if code == -1: print("运动到拍照位置失败") # time.sleep(3) #拍照 rgb, depth, intrinsics = self.cam.get_latest_frame() cv2.imshow("rgb2",rgb) cv2.waitKey(0) if self.check_corners_add_sample() is False: print("初始位置下没有找到标定板") rgb, depth, intrinsics = self.cam.get_latest_frame() pose = self.arm.robot_rpc_client.getRobotInterface(self.arm.robot_name).getRobotState().getTcpPose() cv2.imshow("not right",rgb) cv2.waitKey(0) rgb, depth, intrinsics = self.cam.get_latest_frame() gray = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY) ret, corners = cv2.findChessboardCorners(gray, self.board_size, None) robot_pose = self.get_pose_homomat() print(f"EE pose:{robot_pose}") center = self.init_chessboard_center(rgb) print(f"current center EE: {center}") ''' Z轴抖动 ''' # EE坐标系下的点 trajectory, rpy_values = self.generate_cyclic_trajectory_above_center(center) self.visualization(center,trajectory,rpy_values) homo_trajectory = np.hstack((trajectory, np.ones((len(trajectory), 1)))) tcp_pose = self.get_pose_homomat() # 把生成的trajectory 移动到 世界坐标下 trajectory = (tcp_pose @ homo_trajectory.T).T[:, :3] rpy_angles = [] for tool_point in trajectory: vector_center_to_tool = tool_point - center roll, pitch, yaw = self.get_roll_pitch_yaw(vector_center_to_tool) rpy_angles.append((roll, pitch, yaw)) rpy_angles = np.array(rpy_angles) pose = self.concatenate_path_and_rpy(trajectory,rpy_angles ) #世界坐标系下的中心 center_homo = np.append(center, 1).reshape(4, 1) tcp_pose = self.get_pose_homomat() center_world_homo = tcp_pose @ center_homo # 4x4 *4x1 =4x1 center_world = center_world_homo[:3].flatten() print(f"world center: {center_world} \n top pose:\n {pose[:5]}") lenOfResult = self.traversal_position_direction(pose) ''' Z轴抖动 ''' assert lenOfResult > 20, "没有收集到足够多的数据来完成精确的标定" def generate_cyclic_trajectory_above_center(self,center, num_steps=12, x_amplitude=0.12, z_amplitude=0.20, frequency=4, min_z_offset=0.60,step_z=0.01): """ 生成所有轨迹点都在中心点上方的周期性波动轨迹。 """ trajectory = [] rpy_values = [] y_constant = center[1] x_values = np.linspace(-x_amplitude, x_amplitude, num_steps) y_values = [center[1]+0.05,center[1]+0.15] # y_values = [center[1]+0.05,center[1]+0.15,center[1]+0.30] # y_values = [center[1]+0.15,center[1]+0.03,center[1]+0.06,center[1]+0.10,center[1]+0.15] # y_values = [center[1],center[1],center[1]] for i in range(len(y_values)): y = y_values[i] for x in x_values: # 使用绝对值正弦函数确保 Z 始终在 center[2] 之上 z = z_amplitude * np.abs(np.sin(frequency * x * np.pi)) + center[2] + min_z_offset + step_z*i tool_point = np.array([x + center[0], y, z]) trajectory.append(tool_point) # 计算从中心点射向轨迹点的向量 vector_center_to_tool = tool_point - center roll, pitch, yaw = self.get_roll_pitch_yaw(vector_center_to_tool) rpy_values.append((roll, pitch, yaw)) print(f"above part center:{center}\n {trajectory[:5]}") return np.array(trajectory), np.array(rpy_values) def get_roll_pitch_yaw(self,vector): """ 计算从原点指向给定向量的 roll, pitch, yaw (弧度). 假设 Z 轴向前，Y 轴向上，X 轴向右。 """ x, y, z = vector roll = math.atan2(y, z) pitch = math.atan2(-x, math.sqrt(y**2 + z**2)) yaw = 0 # 假设指向 tool 的方向没有偏航，或者根据实际需求调整 return roll, pitch, yaw def concatenate_path_and_rpy(self,path, rpy_angles): """ Concatenates a list of positions [x, y, z] with a corresponding list of RPY angles [roll, pitch, yaw]. Args: path (list of lists): A list where each inner list contains [x, y, z]. rpy_angles (list of lists or tuples): A list where each inner list or tuple contains [roll, pitch, yaw]. Returns: list of lists: A new list where each inner list is a concatenation of the corresponding position and RPY angles: [x, y, z, roll, pitch, yaw]. """ if len(path) != len(rpy_angles): raise ValueError("The lengths of the path and rpy_angles lists must be the same.") combined_list = [] for i in range(len(path)): position = path[i] orientation = rpy_angles[i] # combined = np.concatenate((position,orientation)) combined = np.hstack((position.flatten(), orientation)) combined_list.append(combined) print(f"combined xyz|rpy :{combined}") return combined_list def check_corners_add_sample(self): rgb, depth, intrinsics = self.cam.get_latest_frame() gray = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY) ret, corners = cv2.findChessboardCorners(gray, self.board_size) pic_name = self.file_address + "/"+str(len(self.image_points)) + ".jpg" if ret: robot_pose = self.get_pose_homomat() self.add_sample(gray,corners,robot_pose) print(f"save path as: {pic_name}") # 绘制检测到的角点 (用于可视化，可选) # cv2.drawChessboardCorners(rgb, self.board_size, corners, ret) cv2.imwrite(pic_name, rgb) return True cv2.imwrite(pic_name, rgb) return False def add_sample(self, gray, corners,robot_pose): """ 添加标定样本。 image: 捕获的图像 robot_pose: 机械臂的位姿，格式为 [x, y, z, roll, pitch, yaw] """ # 检测棋盘角点 corners2 = cv2.cornerSubPix(gray, corners, (11, 11), (-1, -1), criteria) # 在原角点的基础上寻找亚像素角点 # if self._validate_corners(corners2): if [corners2]: self.obj_points.append(self.objp) self.image_points.append(corners2) self.robot_poses.append(robot_pose) print(f"insert {robot_pose}") def get_pose_homomat(self): pose = self.arm.robot_rpc_client.getRobotInterface(self.arm.robot_name).getRobotState().getTcpPose() x, y, z, rx, ry, rz = pose r = R.from_euler('xyz', [rx, ry, rz]) rotation_matrix = r.as_matrix() translation_vector = np.array([x, y, z]).reshape(3, 1) homo_mat = np.eye(4) # 创建 4x4 单位矩阵 homo_mat[:3, :3] = rotation_matrix homo_mat[:3, 3] = translation_vector.flatten() return homo_mat def camera_calib(self): ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(self.obj_points, self.image_points, (self.size[1],self.size[0]) , None, None) if ret: print("内参矩阵:\n", mtx) # 内参数矩阵 print("畸变系数:\n", dist) # 畸变系数 distortion cofficients = (k_1,k_2,p_1,p_2,k_3) print("++++++++++相机标定完成++++++++++++++") else: print("-----------相机标定失败--------------") return rvecs, tvecs def split_robot_pose(self): """ 将包含 4x4 齐次矩阵的 robot_pose 列表拆分成 R_arm 和 t_arm。 Args: robot_pose: 包含 4x4 齐次矩阵的 Python list。 Returns: R_arm: 旋转矩阵的 Python list, 每个元素是 3x3 numpy 数组。 t_arm: 平移向量的 Python list, 每个元素是 3x1 numpy 数组。 """ R_arm = [] t_arm = [] for pose_matrix in self.robot_poses: # 1. 提取旋转矩阵 (3x3) rotation_matrix = pose_matrix[:3, :3] R_arm.append(rotation_matrix) # 2. 提取平移向量 (3x1) translation_vector = pose_matrix[:3, 3] # 获取齐次矩阵的平移部分 (1x3 numpy 数组) translation_vector = translation_vector.reshape(3, 1) # 转换为 3x1 的列向量 t_arm.append(translation_vector) return R_arm, t_arm def calibrate(self): rvecs, tvecs = self.camera_calib() r_arm, t_arm= self.split_robot_pose() rvecs_array = np.array(rvecs) tvecs_array = np.array(tvecs) t_arma = np.array(t_arm) r_arma = np.array(r_arm) print(rvecs_array.shape) print(tvecs_array.shape) print(t_arma.shape) print(r_arma.shape) R, t = cv2.calibrateHandEye(r_arma, t_arma, rvecs_array, tvecs_array, cv2.CALIB_HAND_EYE_HORAUD) print("+++++++++++手眼标定完成+++++++++++++++") return R, t def init_chessboard_center(self,image): ''' 返回的是EE坐标下的中点 ''' intrinsics = self.intrinsics # 检测标定板获取初始位置 ret, corners = cv2.findChessboardCorners(image, self.board_size) if not ret: raise ValueError("标定板未检测到") # 计算标定板中心在相机坐标系中的位置 obj_pts = np.array([[i*self.square_size, j*self.square_size, 0] for i in range(self.board_size[0]) for j in range(self.board_size[1])], dtype=np.float32) print(intrinsics) camera_matrix = np.array([ [intrinsics['fx'], 0, intrinsics['cx']], [0, intrinsics['fy'], intrinsics['cy']], [0, 0, 1] ], dtype=np.float32) dist_coeffs = np.array([ intrinsics['distortion_coeffs']['k1'], intrinsics['distortion_coeffs']['k2'], intrinsics['distortion_coeffs']['p1'], intrinsics['distortion_coeffs']['p2'], intrinsics['distortion_coeffs']['k3'] ], dtype=np.float32) # 同样确保数据类型 print("obj_pts shape:", obj_pts.shape) print("corners shape:", corners.shape) ret, rvec, tvec = cv2.solvePnP(obj_pts, corners, camera_matrix, dist_coeffs) # 这里我们假设标定板的初始位置是原点，没有旋转和平移 rvec_initial = np.zeros((3, 1), dtype=np.float32) # 初始旋转向量 tvec_initial = np.zeros((3, 1), dtype=np.float32) # 初始平移向量 # 初始变换（标定板到初始坐标系） initial_transform = np.eye(4) initial_transform[:3, :3], _ = cv2.Rodrigues(rvec_initial) initial_transform[:3, 3] = tvec_initial.flatten() print(f"board itselt {initial_transform}") # PnP 结果（初始坐标系到相机） cam2initial = np.eye(4) cam2initial[:3, :3], _ = cv2.Rodrigues(rvec) cam2initial[:3, 3] = tvec.flatten() # 合成最终变换：标定板 -> 初始坐标系 -> 相机 board2cam = initial_transform @ cam2initial print(f"board2cam\n{board2cam}") # 手眼标定矩阵的逆（相机到TCP的固定变换） cam2tool = np.eye(4) cam2tool[:3, :3] = np.array([[0.0, 1.0 , 0.0], [1.0 , 0.0 , 0.0], [0.0, 0.0 , -1.0]]) cam2tool[:3, 3] = np.array([0.00209053, 0.01021561, -0.16877656]).flatten() ###找到棋盘格中心 camera_center = None chessboard_center_object = self.get_chessboard_center_object_points() if chessboard_center_object is not None: # Convert to homogeneous coordinates chessboard_center_homogeneous = np.append(chessboard_center_object, 1) # Transform the center to camera coordinates camera_center_homogeneous = cam2tool@ board2cam @ chessboard_center_homogeneous camera_center = camera_center_homogeneous[:3] return camera_center def get_chessboard_center_object_points(self): rows, cols = self.board_size obj_pts = self.objp if obj_pts is None or len(obj_pts) != rows * cols: return None # Get the range of x and y indices x_indices = obj_pts[:, 0].flatten() y_indices = obj_pts[:, 1].flatten() center_x_index = (np.max(x_indices) + np.min(x_indices)) / 2.0 center_y_index = (np.max(y_indices) + np.min(y_indices)) / 2.0 center_z = 0 # Assuming the chessboard is on the Z=0 plane in its frame return np.array([center_x_index, center_y_index, center_z]) def generate_perturbed_joint_angles_rad(self,original_angle): """ Args: original_angle (list or tuple): 包含6个角度值的列表或元组 (单位：度). Returns: numpy.ndarray: 一个形状为 (6, 6) 的 numpy 数组，包含6组6个角度值（单位：弧度）。 每组数据的含义如下： - 第1组: 原始角度（转换为弧度）。 - 第2组: 第五个关节角度 + rand5（转换为弧度）。 - 第3组: 第五个关节角度 - rand5（转换为弧度）。 - 第4组: 第四个关节角度 + rand4（转换为弧度）。 - 第5组: 第四个关节角度 - rand4（转换为弧度）。 - 第6组: 第四个关节角度 + rand4，第五个关节角度 + rand5（转换为弧度）。 """ original_angle = np.array(original_angle) if original_angle.shape != (6,): raise ValueError("输入的角度列表必须包含6个元素。") original_angle[5] = 0.0 # 生成随机数 rand5 = random.uniform(-20*np.pi/180, 20*np.pi/180) # 第五个关节的随机数，范围可以调整 rand4 = random.uniform(-10*np.pi/180, 10*np.pi/180) # 第四个关节的随机数，范围可以调整 angle_array = [] # 1. 原始角度 angle_array.append(original_angle.copy()) # 4. += rand4 angle_c = original_angle.copy() angle_c[3] -= rand4 angle_c[4] -= rand5 angle_array.append(angle_c) # 5. -= rand4 angle_d = original_angle.copy() angle_d[3] += rand4 angle_d[4] += rand5 angle_array.append(angle_d) # 6. +=rand4，-= rand5 angle_e = original_angle.copy() angle_e[3] += rand4 angle_e[4] -= rand5 angle_array.append(angle_e) # 6. -= rand4，第五个关节 += rand5 angle_f = original_angle.copy() angle_f[3] -= rand4 angle_f[4] += rand5 angle_array.append(angle_f) # 6. -= rand4，第五个关节 += rand5 angle_g = original_angle.copy() angle_g[3] -= rand4/2 angle_g[4] += rand5/2 angle_array.append(angle_g) angle_h = original_angle.copy() angle_h[3] += rand4/2 angle_h[4] -= rand5/2 angle_array.append(angle_h) # 将所有角度从度转换为弧度 # modified_angles_rad = np.array([np.deg2rad(angles) for angles in angle_array]) # import pdb # pdb.set_trace() return angle_array def traversal_position_direction(self,path): for pose in path: current_pose = self.arm.robot_rpc_client.getRobotInterface(self.arm.robot_name).getRobotState().getTcpPose() print(f" elipse path is {pose}\n current pose is {current_pose}" ) code = self.arm.move_line( pose ) if code == -1: print(f"运动到拍照位置失败:{pose}") else: joint = self.arm.robot_rpc_client.getRobotInterface(self.arm.robot_name).getRobotState().getJointPositions() angle_list = self.generate_perturbed_joint_angles_rad(joint) print(f"\n {angle_list}\n Angles Position ") for joint in angle_list: print(f"current joint: {joint}") code = self.arm.move_joint(joint,4,wait=True) if code != -1: print(f"move success") self.check_corners_add_sample() return len(self.robot_poses) def visualization(self,center,trajectory,rpy_values): # 可视化 fig = plt.figure(figsize=(10, 8)) ax = fig.add_subplot(111, projection='3d') # 绘制中心点 ax.scatter(*center, color='r', marker='o', s=100, label='Center Point') # 绘制轨迹 ax.plot(trajectory[:, 0], trajectory[:, 1], trajectory[:, 2], color='b', label='Tool Trajectory') # 可视化 RPY 值 (每隔几个点显示一个) num_samples_rpy = 20 indices = np.linspace(0, len(rpy_values) - 1, num_samples_rpy, dtype=int) for i in indices: tool_point = trajectory[i] roll, pitch, yaw = rpy_values[i] vector = tool_point - center # 将箭头的起始点移动到 tool_point ax.quiver(tool_point[0], tool_point[1], tool_point[2], vector[0], vector[1], vector[2], length=0.2, color='g', alpha=0.7, label='Vector Center to Tool' if i == indices[0] else "") # 将文本标签的位置移动到 tool_point 附近 ax.text(tool_point[0] + 0.05, tool_point[1] + 0.05, tool_point[2] + 0.05, f'RPY:({math.degrees(roll):.2f},{math.degrees(pitch):.2f},{math.degrees(yaw):.2f})', color='g', fontsize=8) ax.set_xlabel('X') ax.set_ylabel('Y') ax.set_zlabel('Z') ax.set_title('Cyclic Trajectory Above Center Point') ax.legend() ax.view_init(azim=-120, elev=30) plt.grid(True) plt.show() # 打印 RPY 值 (示例) # print("\nRoll, Pitch, Yaw values (degrees) along the trajectory:") # for i, (r, p, y) in enumerate(rpy_values): # if i % 10 == 0: # Print every 10th value # print(f"Step {i}: Roll: {math.degrees(r):.2f}, Pitch: {math.degrees(p):.2f}, Yaw: {math.degrees(y):.2f}") if __name__ == '__main__': # 标定板的width 对应的角点的个数 6 # 标定板的height 对应的角点的个数 3 calibration_board_size = [9,6] calibration_board_square_size = 0.020 #unit - meter systemPath = "/home/jsfb/Documents/" now = datetime.now() date_time_str = now.strftime("%d_%H_%M") folder_name = f"calibration_data_{date_time_str}" systemPath+=folder_name os.makedirs(systemPath, exist_ok=True) calib = Calibration(calibration_board_size,calibration_board_square_size,systemPath) calib.collect_data() r, t = calib.calibrate() print("旋转矩阵：") print(r) print("平移向量：") print(t) # arm = AuboC5() # arm.init() # aubo_algorithm = arm.robot_rpc_client.getRobotInterface(arm.robot_name).getRobotAlgorithm().inverseKinematics() # # 查看 rpc_client 对象的所有属性和方法 # print(dir(aubo_algorithm))