File size: 8,383 Bytes
dcb4bf3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 |
import os
import time
from turtle import pos
import numpy as np
import pybullet as p
from surrol.const import ASSET_DIR_PATH
from surrol.tasks.psm_env import PsmEnv, goal_distance
from surrol.utils.pybullet_utils import get_link_pose, wrap_angle
class PegTransfer(PsmEnv):
POSE_BOARD = ((0.55, 0, 0.6861), (0, 0, 0)) # 0.675 + 0.011 + 0.001
WORKSPACE_LIMITS = ((0.50, 0.60), (-0.05, 0.05), (0.686, 0.745))
SCALING = 5.
# TODO: grasp is sometimes not stable; check how to fix it
def _env_setup(self):
super(PegTransfer, self)._env_setup()
self.has_object = True
# robot
workspace_limits = self.workspace_limits1
pos = (workspace_limits[0][0],
workspace_limits[1][1],
workspace_limits[2][1])
orn = (0.5, 0.5, -0.5, -0.5)
joint_positions = self.psm1.inverse_kinematics((pos, orn), self.psm1.EEF_LINK_INDEX)
self.psm1.reset_joint(joint_positions)
self.block_gripper = False
# peg board
obj_id = p.loadURDF(os.path.join(ASSET_DIR_PATH, 'peg_board/peg_board.urdf'),
np.array(self.POSE_BOARD[0]) * self.SCALING,
p.getQuaternionFromEuler(self.POSE_BOARD[1]),
globalScaling=self.SCALING)
self.obj_ids['fixed'].append(obj_id) # 1
self._pegs = np.arange(12)
np.random.shuffle(self._pegs[:6])
np.random.shuffle(self._pegs[6: 12])
#self._pegs = [ 1 , 2 , 3 , 4 , 5 , 0 , 7 ,11 , 9 ,10 , 6 , 8]
# blocks
num_blocks = 4
# for i in range(6, 6 + num_blocks):
for i in self._pegs[6: 6 + num_blocks]:
pos, orn = get_link_pose(self.obj_ids['fixed'][1], i)
yaw = (np.random.rand() - 0.5) * np.deg2rad(60)
obj_id = p.loadURDF(os.path.join(ASSET_DIR_PATH, 'block/block.urdf'),
np.array(pos) + np.array([0, 0, 0.03]),
p.getQuaternionFromEuler((0, 0, yaw)),
useFixedBase=False,
globalScaling=self.SCALING)
self.obj_ids['rigid'].append(obj_id)
self._blocks = np.array(self.obj_ids['rigid'][-num_blocks:])
np.random.shuffle(self._blocks)
#self._blocks = [5,8,7,6]
for obj_id in self._blocks[:1]:
# change color to red
p.changeVisualShape(obj_id, -1, rgbaColor=(255 / 255, 69 / 255, 58 / 255, 1))
self.obj_id, self.obj_link1 = self._blocks[0], 1
#print(self._pegs, self.obj_id, self._blocks)
def _is_success(self, achieved_goal, desired_goal):
""" Indicates whether or not the achieved goal successfully achieved the desired goal.
"""
# TODO: may need to tune parameters
return np.logical_and(
goal_distance(achieved_goal[..., :2], desired_goal[..., :2]) < 5e-3 * self.SCALING,
np.abs(achieved_goal[..., -1] - desired_goal[..., -1]) < 4e-3 * self.SCALING
).astype(np.float32)
def _sample_goal(self) -> np.ndarray:
""" Samples a new goal and returns it.
"""
goal = np.array(get_link_pose(self.obj_ids['fixed'][1], self._pegs[0])[0])
return goal.copy()
def _sample_goal_callback(self):
""" Define waypoints
"""
super()._sample_goal_callback()
self._waypoints = [None, None, None, None, None, None, None] # six waypoints
pos_obj, orn_obj = get_link_pose(self.obj_id, self.obj_link1)
orn = p.getEulerFromQuaternion(orn_obj)
orn_eef = get_link_pose(self.psm1.body, self.psm1.EEF_LINK_INDEX)[1]
orn_eef = p.getEulerFromQuaternion(orn_eef)
yaw = orn[2] if abs(wrap_angle(orn[2] - orn_eef[2])) < abs(wrap_angle(orn[2] + np.pi - orn_eef[2])) \
else wrap_angle(orn[2] + np.pi) # minimize the delta yaw
self._waypoints[0] = np.array([pos_obj[0], pos_obj[1],
pos_obj[2] + 0.045 * self.SCALING, yaw, 0.5]) # above object
self._waypoints[1] = np.array([pos_obj[0], pos_obj[1],
pos_obj[2] + (0.003 + 0.0102) * self.SCALING, yaw, 0.5]) # approach
self._waypoints[2] = np.array([pos_obj[0], pos_obj[1],
pos_obj[2] + (0.003 + 0.0102) * self.SCALING, yaw, -0.5]) # grasp
self._waypoints[3] = np.array([pos_obj[0], pos_obj[1],
pos_obj[2] + 0.045 * self.SCALING, yaw, -0.5]) # lift up
# pos_peg = get_link_pose(self.obj_ids['fixed'][1], self.obj_id - np.min(self._blocks) + 6)[0] # 6 pegs
pos_peg = get_link_pose(self.obj_ids['fixed'][1],
self._pegs[self.obj_id - np.min(self._blocks) + 6])[0] # 6 pegs
pos_place = [self.goal[0] + pos_obj[0] - pos_peg[0],
self.goal[1] + pos_obj[1] - pos_peg[1], self._waypoints[0][2]] # consider offset
self._waypoints[4] = np.array([pos_place[0], pos_place[1], pos_place[2], yaw, -0.5]) # above goal
self._waypoints[5] = np.array([pos_place[0], pos_place[1], self._waypoints[2][2], yaw, -0.5]) # release
#self._waypoints[5] = np.array([pos_place[0], pos_place[1], pos_place[2], yaw, 0.5]) # release
self._waypoints[6] = np.array([pos_place[0], pos_place[1], self._waypoints[2][2], yaw, 0.5]) # release
self.waypoints = self._waypoints.copy()
def _meet_contact_constraint_requirement(self):
# add a contact constraint to the grasped block to make it stable
pose = get_link_pose(self.obj_id, -1)
return pose[0][2] > self.goal[2] + 0.01 * self.SCALING
def get_oracle_action(self, obs) -> np.ndarray:
"""
Define a human expert strategy
"""
# six waypoints executed in sequential order
action = np.zeros(5)
for i, waypoint in enumerate(self._waypoints):
if waypoint is None:
continue
# if i == 3:
# delta_pos = (waypoint[:3] - obs['achieved_goal'][:3]) / 0.01 / self.SCALING
# else:
delta_pos = (waypoint[:3] - obs['observation'][:3]) / 0.01 / self.SCALING
delta_yaw = (waypoint[3] - obs['observation'][5]).clip(-0.4, 0.4)
if np.abs(delta_pos).max() > 1:
delta_pos /= np.abs(delta_pos).max()
scale_factor = 0.7
delta_pos *= scale_factor
action = np.array([delta_pos[0], delta_pos[1], delta_pos[2], delta_yaw, waypoint[4]])
if np.linalg.norm(delta_pos) * 0.01 / scale_factor < 2e-3 and np.abs(delta_yaw) < np.deg2rad(2.):
self._waypoints[i] = None
break
return action
def subgoal_grasp(self):
scale_factor = 0.7
# robot_state
robot_state = self._get_robot_state(idx=0)
pos_robot = robot_state[:3]
yaw_robot = robot_state[5]
orn_eef = get_link_pose(self.psm1.body, self.psm1.EEF_LINK_INDEX)[1]
orn_eef = p.getEulerFromQuaternion(orn_eef)
yaw_angle = robot_state[-1]
pos_tip = np.array(get_link_pose(self.psm1.body, self.psm1.TIP_LINK_INDEX)[0])
# object pose
pos_obj, orn_obj = get_link_pose(self.obj_id, self.obj_link1)
orn = p.getEulerFromQuaternion(orn_obj)
yaw_grasp = orn[2] if abs(wrap_angle(orn[2] - orn_eef[2])) < abs(wrap_angle(orn[2] + np.pi - orn_eef[2])) \
else wrap_angle(orn[2] + np.pi) # minimize the delta yaw
pos_grasp = np.array([pos_obj[0], pos_obj[1],
pos_obj[2]]) # grasp
# is_success
is_grasp = yaw_angle < 0.2 # TODO: fine-tune
is_success = self._is_success(pos_grasp, pos_tip) #and (is_grasp or self._meet_contact_constraint_requirement())
return is_success.astype(np.float32)
def is_success(self, achieved_goal, desired_goal):
""" Indicates whether or not the achieved goal successfully achieved the desired goal.
"""
return self._is_success(achieved_goal, desired_goal)
@property
def pegs(self):
return self._pegs
if __name__ == "__main__":
env = PegTransfer(render_mode='human') # create one process and corresponding env
env.test()
env.close()
time.sleep(2)
|