# Author(s): Jiaqi Xu # Created on: 2020-11 """ PSM wrapper Refer to: https://github.com/jhu-dvrk/dvrk-ros/blob/master/dvrk_python/src/dvrk/ecm.py https://github.com/jhu-dvrk/dvrk-ros/blob/7b3d48ca164755ccfc88028e15baa9fbf7aa1360/dvrk_python/src/dvrk/ecm.py https://github.com/jhu-dvrk/sawIntuitiveResearchKit/blob/master/share/kinematic/ecm.json https://github.com/jhu-dvrk/sawIntuitiveResearchKit/blob/4a8b4817ee7404b3183dfba269c0efe5885b41c2/share/arm/ecm-straight.json """ import os import numpy as np import pybullet as p from surrol.robots.arm import Arm from surrol.const import ASSET_DIR_PATH from surrol.utils.pybullet_utils import ( get_joint_positions, get_link_pose, render_image ) # Rendering width and height RENDER_HEIGHT = 256 RENDER_WIDTH = 256 FoV = 60 LINKS = ( 'ecm_base_link', 'ecm_yaw_link', 'ecm_pitch_end_link', # -1, 0, 1 'ecm_main_insertion_link', 'ecm_tool_link', # 2, 3 'ecm_end_link', # 4 'ecm_tip_link', # 5 'ecm_pitch_front_link', # 6 'ecm_pitch_bottom_link', 'ecm_pitch_top_link', # 7, 8 'ecm_pitch_back_link', # 9 'ecm_remote_center_link', # 10 ) # tooltip-offset; refer to .json tool_T_tip = np.array([[0.0, 1.0, 0.0, 0.0], [-1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 0.0, 1.0]]) # Joint limits. No limits in the .json. TODO: dVRK config modified TOOL_JOINT_LIMIT = { 'lower': np.deg2rad([-90.0, -45.0, 0.0, -np.inf]), # not sure about the last joint 'upper': np.deg2rad([ 90.0, 66.4, 254.0, np.inf]), } TOOL_JOINT_LIMIT['lower'][2] = -0.01 # allow small tolerance TOOL_JOINT_LIMIT['upper'][2] = 0.254 # prismatic joint (m); not sure, from ambf # [-1.57079633, -0.78539816, 0. , -1.57079633] # [ 1.57079633, 1.15889862, 0.254, 1.57079633] class Ecm(Arm): NAME = 'ECM' URDF_PATH = os.path.join(ASSET_DIR_PATH, 'ecm/ecm.urdf') DoF = 4 # 4-dof arm JOINT_TYPES = ('R', 'R', 'P', 'R') EEF_LINK_INDEX = 4 # EEF link index, one redundant joint for inverse kinematics TIP_LINK_INDEX = 5 # redundant joint for easier camera matrix computation RCM_LINK_INDEX = 10 # RCM link index # D-H parameters A = np.array([0.0, 0.0, 0.0, 0.0]) ALPHA = np.array([np.pi / 2, -np.pi / 2, np.pi / 2, 0.0]) D = np.array([0.0, 0.0, -0.3822, 0.3829]) THETA = np.array([np.pi / 2, -np.pi / 2, 0.0, 0.0]) def __init__(self, pos=(0., 0., 1.), orn=(0., 0., 0., 1.), scaling=1.): super(Ecm, self).__init__(self.URDF_PATH, pos, orn, TOOL_JOINT_LIMIT, tool_T_tip, scaling) # camera control related parameters self.view_matrix = None self.proj_matrix = None self._homo_delta = np.zeros((2, 1)) self._wz = 0 # b: rcm, e: eef, c: camera pos_eef, orn_eef = get_link_pose(self.body, self.EEF_LINK_INDEX) pos_cam, orn_cam = get_link_pose(self.body, self.TIP_LINK_INDEX) self._tip_offset = np.linalg.norm(np.array(pos_eef) - np.array(pos_cam)) # TODO wRe = np.array(p.getMatrixFromQuaternion(orn_eef)).reshape((3, 3)) wRc = np.array(p.getMatrixFromQuaternion(orn_cam)).reshape((3, 3)) self._wRc0 = wRc.copy() # initial rotation matrix self._eRc = np.matmul(wRe.T, wRc) def _get_joint_positions_all(self, abs_input): """ With the consideration of parallel mechanism constraints and other redundant joints. """ positions = get_joint_positions(self.body, self.joints) joint_positions = [ abs_input[0], abs_input[1], # 0, 1 abs_input[2] * self.scaling, abs_input[3], # 2, 3 positions[4], positions[5], # 4 (0.0), 5 (0.0) abs_input[1], # 6 -abs_input[1], -abs_input[1], # 7, 8 abs_input[1], # 9 positions[10], # 10 (0.0) ] return joint_positions def cVc_to_dq(self, cVc: np.ndarray) -> np.ndarray: """ convert the camera velocity in its own frame (cVc) into the joint velocity q_dot """ cVc = cVc.reshape((3, 1)) # restrict the step size, need tune if np.abs(cVc).max() > 0.01: cVc = cVc / np.abs(cVc).max() * 0.01 # Forward kinematics q = self.get_current_joint_position() bRe = self.robot.fkine(q).R # use rtb instead of PyBullet, no tool_tip_offset _, orn_cam = get_link_pose(self.body, self.TIP_LINK_INDEX) wRc = np.array(p.getMatrixFromQuaternion(orn_cam)).reshape((3, 3)) # Rotation R1, R2 = self._wRc0, wRc x = R1[0, 0] * R2[1, 0] - R1[1, 0] * R2[0, 0] + R1[0, 1] * R2[1, 1] - R1[1, 1] * R2[0, 1] y = R1[0, 0] * R2[1, 1] - R1[1, 0] * R2[0, 1] - R1[0, 1] * R2[1, 0] + R1[1, 1] * R2[0, 0] dz = np.arctan(x / y) k1, k2 = 25.0, 0.1 self._wz = k1 * dz * np.exp(-k2 * np.linalg.norm(self._homo_delta)) # print(' -> x: {:.4f}, y: {:.4f}, dz: {:.4f}, wz: {:.4f}'.format(x, y, dz, self._wz)) # Pseudo Solution d = self._tip_offset Jd = np.matmul(self._eRc, np.array([[0, 0, d, 0], [0, -d, 0, 0], [1, 0, 0, 0]])) Je = np.matmul(self._eRc, np.array([[0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]])) eVe4 = np.dot(np.linalg.pinv(Jd), cVc) \ + np.dot(np.dot((np.eye(4) - np.dot(np.linalg.pinv(Jd), Jd)), np.linalg.pinv(Je)), np.array([[0], [0], [self._wz]])) eVe = np.zeros((6, 1)) eVe[2: 6] = eVe4[0: 4] Q = np.zeros((6, 6)) Q[0: 3, 0: 3] = - bRe Q[3: 6, 3: 6] = - bRe bVe = np.dot(Q, eVe) # Compute the Jacobian matrix bJe = self.get_jacobian_spatial() dq = np.dot(np.linalg.pinv(bJe), bVe) # print(" -> cVc: {}, q: {}, dq: {}".format(list(np.round(cVc.flatten(), 4)), q, list(dq.flatten()))) return dq.flatten() def render_image(self, width=RENDER_WIDTH, height=RENDER_HEIGHT): pos_eef, orn_eef = get_link_pose(self.body, self.EEF_LINK_INDEX) pos_tip = get_link_pose(self.body, self.TIP_LINK_INDEX)[0] mat_eef = np.array(p.getMatrixFromQuaternion(orn_eef)).reshape((3, 3)) # TODO: need to check the up vector self.view_matrix = p.computeViewMatrix(cameraEyePosition=pos_eef, cameraTargetPosition=pos_tip, cameraUpVector=mat_eef[:, 0]) self.proj_matrix = p.computeProjectionMatrixFOV(fov=FoV, aspect=float(width) / height, nearVal=0.01, farVal=10.0) rgb_array, mask, depth = render_image(width, height, self.view_matrix, self.proj_matrix) return rgb_array, mask, depth def get_centroid_proj(self, pos) -> np.ndarray: """ Compute the object position in the camera NDC space. Refer to OpenGL. :param pos: object position in the world frame. """ assert len(pos) in (3, 4) if len(pos) == 3: # homogeneous coordinates: (x, y, z) -> (x, y, z, w) pos_obj = np.ones((4, 1)) pos_obj[: 3, 0] = pos else: pos_obj = np.array(pos).reshape((4, 1)) view_matrix = np.array(self.view_matrix).reshape(4, 4).T proj_matrix = np.array(self.proj_matrix).reshape(4, 4).T # pos in the camera frame pos_cam = np.dot(proj_matrix, np.dot(view_matrix, pos_obj)) pos_cam /= pos_cam[3, 0] return np.array([pos_cam[0][0], - pos_cam[1][0]]) # be consistent with get_centroid @property def homo_delta(self): return self._homo_delta @homo_delta.setter def homo_delta(self, val: np.ndarray): self._homo_delta = val @property def wz(self): return self._wz