from typing import Any, Dict, Union import numpy as np import sapien import torch from mani_skill.agents.robots import SO100, Fetch, Panda from mani_skill.envs.sapien_env import BaseEnv from mani_skill.envs.tasks.tabletop.pick_cube_cfgs import PICK_CUBE_CONFIGS from mani_skill.sensors.camera import CameraConfig from mani_skill.utils import sapien_utils from mani_skill.utils.building import actors from mani_skill.utils.registration import register_env from mani_skill.utils.scene_builder.table import TableSceneBuilder from mani_skill.utils.structs.pose import Pose from mani_skill.utils import common, sapien_utils from mani_skill.utils.structs import Actor #Robomme import matplotlib.pyplot as plt from mani_skill.utils.geometry.rotation_conversions import ( euler_angles_to_matrix, matrix_to_quaternion, ) import copy from .utils import * from .utils.difficulty import normalize_robomme_difficulty from .utils.subgoal_evaluate_func import static_check from .utils import subgoal_language from .utils.object_generation import spawn_fixed_cube, build_board_with_hole from .utils import reset_panda from ..logging_utils import logger PICK_CUBE_DOC_STRING = """**Task Description:** A simple task where the objective is to grasp a red cube with the {robot_id} robot and move it to a target goal position. This is also the *baseline* task to test whether a robot with manipulation capabilities can be simulated and trained properly. Hence there is extra code for some robots to set them up properly in this environment as well as the table scene builder. **Randomizations:** - the cube's xy position is randomized on top of a table in the region [0.1, 0.1] x [-0.1, -0.1]. It is placed flat on the table - the cube's z-axis rotation is randomized to a random angle - the target goal position (marked by a green sphere) of the cube has its xy position randomized in the region [0.1, 0.1] x [-0.1, -0.1] and z randomized in [0, 0.3] **Success Conditions:** - the cube position is within `goal_thresh` (default 0.025m) euclidean distance of the goal position - the robot is static (q velocity < 0.2) """ @register_env("MoveCube") class MoveCube(BaseEnv): _sample_video_link = "https://github.com/haosulab/ManiSkill/raw/main/figures/environment_demos/PickCube-v1_rt.mp4" SUPPORTED_ROBOTS = [ "panda", "fetch", "xarm6_robotiq", "so100", "widowxai", ] agent: Union[Panda] goal_thresh = 0.025 cube_spawn_half_size = 0.05 cube_spawn_center = (0, 0) _clearance = 0.01 def __init__(self, *args, robot_uids="panda_wristcam", robot_init_qpos_noise=0,seed=0,Robomme_video_episode=None,Robomme_video_path=None, **kwargs): self.reset_in_proecess=False self.robot_init_qpos_noise = robot_init_qpos_noise if robot_uids in PICK_CUBE_CONFIGS: cfg = PICK_CUBE_CONFIGS[robot_uids] else: cfg = PICK_CUBE_CONFIGS["panda"] self.cube_half_size = cfg["cube_half_size"] self.goal_thresh = cfg["goal_thresh"] self.cube_spawn_half_size = cfg["cube_spawn_half_size"] self.cube_spawn_center = cfg["cube_spawn_center"] self.max_goal_height = cfg["max_goal_height"] self.sensor_cam_eye_pos = cfg["sensor_cam_eye_pos"] self.sensor_cam_target_pos = cfg["sensor_cam_target_pos"] self.human_cam_eye_pos = cfg["human_cam_eye_pos"] self.human_cam_target_pos = cfg["human_cam_target_pos"] self.seed = seed self._hb_generator = torch.Generator() self._hb_generator.manual_seed(int(self.seed)) self.robomme_failure_recovery = bool( kwargs.pop("robomme_failure_recovery", False) ) self.robomme_failure_recovery_mode = kwargs.pop( "robomme_failure_recovery_mode", None ) if isinstance(self.robomme_failure_recovery_mode, str): self.robomme_failure_recovery_mode = ( self.robomme_failure_recovery_mode.lower() ) normalized_robomme_difficulty = normalize_robomme_difficulty( kwargs.pop("difficulty", None) ) if normalized_robomme_difficulty is not None: self.difficulty = normalized_robomme_difficulty else: seed_mod = seed % 3 if seed_mod == 0: self.difficulty = "easy" elif seed_mod == 1: self.difficulty = "medium" else: self.difficulty = "hard" self.restore_flag=False self.use_demonstrationwrapper=False self.demonstration_record_traj=False super().__init__(*args, robot_uids=robot_uids, **kwargs) @property def _default_sensor_configs(self): pose = sapien_utils.look_at( eye=self.sensor_cam_eye_pos, target=self.sensor_cam_target_pos ) camera_eye=[0.3,0,0.4] camera_target =[0,0,-0.2] pose = sapien_utils.look_at( eye=camera_eye, target=camera_target ) return [CameraConfig("base_camera", pose, 256, 256, np.pi / 2, 0.01, 100)] @property def _default_human_render_camera_configs(self): pose = sapien_utils.look_at( eye=self.human_cam_eye_pos, target=self.human_cam_target_pos ) return CameraConfig("render_camera", pose, 512, 512, 1, 0.01, 100) def _load_agent(self, options: dict): super()._load_agent(options, sapien.Pose(p=[-0.615, 0, 0])) def _load_scene(self, options: dict): self.table_scene = TableSceneBuilder( self, robot_init_qpos_noise=0 ) self.table_scene.build() length_tensor = torch.rand(1, generator=self._hb_generator) radius_tensor = torch.rand(1, generator=self._hb_generator) self.length = (0.1 + (0.05 - 0.05) * length_tensor).item() self.radius = (0.01 + (0.005 - 0.005) * radius_tensor).item() # Create a single peg #peg_spawn_translation = np.array([self.length / 2, 0.0, self.radius], dtype=np.float32) base_y = -0.2 if torch.rand(1, generator=self._hb_generator).item() < 0.5 else 0.2 peg_spawn_translation = np.array([0.0, base_y, 0.0], dtype=np.float32) # Generate [-0.05, 0.05] random offset (using torch generator) x_jitter = (torch.rand(1, generator=self._hb_generator).item() - 0.5) * 0.1 y_jitter = (torch.rand(1, generator=self._hb_generator).item() - 0.5) * 0.1 # Apply offset peg_spawn_translation[:2] += np.array([x_jitter, y_jitter], dtype=np.float32) self.peg1_basex=peg_spawn_translation[0] self.peg1_basey=peg_spawn_translation[1] initial_yaw = torch.rand(1, generator=self._hb_generator).item() * (np.pi / 2) - (np.pi / 4) yaw_angles = torch.tensor([[0.0, 0.0, initial_yaw]], dtype=torch.float32) yaw_matrix = euler_angles_to_matrix(yaw_angles, convention="XYZ") yaw_quat = matrix_to_quaternion(yaw_matrix)[0].detach().cpu().numpy().tolist() peg_initial_pose = sapien.Pose( p=peg_spawn_translation.tolist(), q=yaw_quat, ) self.peg, self.peg_head, self.peg_tail = build_peg( self, length=self.length, radius=self.radius, initial_pose=peg_initial_pose, name='peg', head_color= "#EC7357", tail_color= "#EC7357", ) # Create lists for backward compatibility self.pegs = [self.peg] self.peg_heads = [self.peg_head] self.peg_tails = [self.peg_tail] # Store initial poses for all pegs self.peg_init_poses = [peg.pose for peg in self.pegs] self.peg_init_pose = self.pegs[0].pose # Keep backward compatibility #generate another set of pose for another reset base_y = -0.2 if torch.rand(1, generator=self._hb_generator).item() < 0.5 else 0.2 peg_spawn_translation = np.array([0.0, base_y, 0.0], dtype=np.float32) x_jitter = (torch.rand(1, generator=self._hb_generator).item() - 0.5) * 0.1 y_jitter = (torch.rand(1, generator=self._hb_generator).item() - 0.5) * 0.1 peg_spawn_translation[:2] += np.array([x_jitter, y_jitter], dtype=np.float32) initial_yaw = torch.rand(1, generator=self._hb_generator).item() * (np.pi / 2) - (np.pi / 4) yaw_angles = torch.tensor([[0.0, 0.0, initial_yaw]], dtype=torch.float32) yaw_matrix = euler_angles_to_matrix(yaw_angles, convention="XYZ") yaw_quat = matrix_to_quaternion(yaw_matrix)[0].detach().cpu().numpy().tolist() self.peg2_basex=peg_spawn_translation[0] self.peg2_basey=peg_spawn_translation[1] peg_initial_pose = sapien.Pose( p=peg_spawn_translation.tolist(), q=yaw_quat, ) self.peg_init_poses_2=[peg_initial_pose] self.finish_return_flag=False # Define task list, each task contains a dictionary with function, name, demonstration flag, and optional failure_func obj_sample = torch.randint(0, 2, (1,), generator=self._hb_generator) self.obj_flag = -1 if obj_sample.item() == 0 else 1 dir_sample = torch.randint(0, 2, (1,), generator=self._hb_generator) #self.direction = -1 if dir_sample.item() == 0 else 1 self.goal_site = spawn_random_target( self, avoid=None, # Use current avoidance list, containing all spawned cubes include_existing=False, # Manually maintain list include_goal=False, # Manually maintain list region_center=[0.0, 0.0], region_half_size=0.15, radius=self.cube_half_size*2, # Use radius instead of half_size thickness=0.005, # target thickness min_gap=self.cube_half_size*1, # Gap requirement same as cube name_prefix=f"goal_site", generator=self._hb_generator ) self.goal_site_2 = spawn_random_target( self, avoid=None, # Use current avoidance list, containing all spawned cubes include_existing=False, # Manually maintain list include_goal=False, # Manually maintain list region_center=[0.0, 0.0], region_half_size=0.1, radius=self.cube_half_size*2, # Use radius instead of half_size thickness=0.005, # target thickness min_gap=self.cube_half_size*1, # Gap requirement same as cube name_prefix=f"goal_site_2", generator=self._hb_generator ) max_cube_spawn_trials = 128 goal_pos = self.goal_site.pose.p goal_xy = np.asarray(goal_pos) goal_xy = np.asarray(goal_xy, dtype=np.float64).reshape(-1)[:2] def _sample_cube_center(required_distance: float): for _ in range(max_cube_spawn_trials): sampled_x = torch.rand(1, generator=self._hb_generator).item() * 0.2 -0.1 #direction = -1.0 if -self.peg1_basey < 0 else 1.0 #sampled_y = torch.rand(1, generator=self._hb_generator).item() * 0.2 * direction sampled_y = torch.rand(1, generator=self._hb_generator).item() * 0.2 -0.1 candidate_xy = np.array([sampled_x, sampled_y], dtype=np.float64) if np.linalg.norm(candidate_xy - goal_xy) > required_distance: return candidate_xy return None cube_center = _sample_cube_center(self.cube_half_size*5) cube_x, cube_y = float(cube_center[0]), float(cube_center[1]) self.cube = spawn_random_cube( self, region_center=[cube_x, cube_y], color=(1, 0, 0, 1), name_prefix="fixed_cube", region_half_size=0.05, generator=self._hb_generator, half_size=self.cube_half_size, ) self.cube_init_pose=self.cube.pose goal_pos = self.goal_site_2.pose.p goal_xy = np.asarray(goal_pos) goal_xy = np.asarray(goal_xy, dtype=np.float64).reshape(-1)[:2] def _sample_cube_center(required_distance: float): for _ in range(max_cube_spawn_trials): sampled_x = torch.rand(1, generator=self._hb_generator).item() * 0.2 -0.1 #direction = -1.0 if -self.peg2_basey < 0 else 1.0 #sampled_y = torch.rand(1, generator=self._hb_generator).item() * 0.2 * direction sampled_y = torch.rand(1, generator=self._hb_generator).item() * 0.2 -0.1 candidate_xy = np.array([sampled_x, sampled_y], dtype=np.float64) if np.linalg.norm(candidate_xy - goal_xy) > required_distance: return candidate_xy return None cube_center = _sample_cube_center(self.cube_half_size*5) cube_x, cube_y = float(cube_center[0]), float(cube_center[1]) self.cube_2 = spawn_random_cube( self, region_center=[cube_x, cube_y], color=(1, 0, 0, 1), name_prefix="fixed_cube_2", region_half_size=0.05, generator=self._hb_generator, half_size=self.cube_half_size, ) self.cube_init_pose_2=self.cube_2.pose #only need the pose! teleport away in goal2_p = np.array(self.goal_site_2.pose.p.detach().cpu().numpy(), dtype=np.float64, copy=True) self.goal_site_2_pose_p = goal2_p goal2_q = np.array(self.goal_site_2.pose.q.detach().cpu().numpy(), dtype=np.float64, copy=True) self.goal_site_2_pose_q = goal2_q goal1_p = np.array(self.goal_site.pose.p.detach().cpu().numpy(), dtype=np.float64, copy=True) self.goal_site_1_pose_p = goal1_p goal1_q = np.array(self.goal_site.pose.q.detach().cpu().numpy(), dtype=np.float64, copy=True) self.goal_site_1_pose_q = goal1_q def _initialize_episode(self, env_idx: torch.Tensor, options: dict): with torch.device(self.device): self.table_scene.initialize(env_idx) if not hasattr(self, "pegs"): return # Initialize all 3 pegs qpos = np.array( [ 0.0, 0, 0, -np.pi * 4 / 8, 0, np.pi * 2 / 4, np.pi / 4, 0.04, 0.04, ], dtype=np.float32, ) self.ways=["peg_push","gripper_push","grasp_putdown"] way_idx = torch.randint(len(self.ways), (1,), generator=self._hb_generator).item() self.way = self.ways[way_idx] #self.way="gripper_push" self.agent.reset(qpos) self.cube_2.set_pose(sapien.Pose(p=[10,10,1]))#only need the pose! self.goal_site_2.set_pose(sapien.Pose(p=[10, -10, 1])) def evaluate(self,solve_complete_eval=False): timestep = self.elapsed_steps # flag=is_A_pickup_notB(self,self.peg_head,self.peg_tail) # flag2=is_A_pickup_notB(self,self.peg_tail,self.peg_head) # flag=is_A_insert_notB(self,self.peg_head,self.peg_tail,self.box) self.successflag=torch.tensor([False]) self.failureflag = torch.tensor([False]) self.obj_flag=-1 if self.obj_flag==-1: self.grasp_target=self.peg_tail self.grasp_target_false=self.peg_head else: self.grasp_target=self.peg_head self.grasp_target_false=self.peg_tail self.direction1 = 1 if self.cube_init_pose.p[0][1]-self.goal_site_1_pose_p[0][1] > 0 else -1# relative position self.direction2 = 1 if self.cube_init_pose_2.p[0][1]-self.goal_site_2_pose_p[0][1] > 0 else -1 # direction -1 push from left # direction 1 push from right +y side / table right side from camera view -> treated as push from right if self.way=="peg_push": tasks = [ { "func": lambda: is_any_obj_pickup_flag_currentpickup(self, objects=[self.grasp_target,self.grasp_target_false]), "name": f"Pick up the peg", "subgoal_segment":f"Pick up the peg at <>", "choice_label": "pick up the peg", "demonstration": True, "failure_func": lambda:[ is_obj_pickup(self, obj=self.cube), is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2),], "solve": lambda env, planner:grasp_and_lift_peg_side(env, planner, env.grasp_target), "segment":self.grasp_target }, { "func": lambda: is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2,must_gripper_open=True), "name": f"Hook the cube to the target with the peg", "subgoal_segment":f"Hook the cube at <> to the target at <> with the peg", "choice_label": "hook the cube to the target with the peg", "demonstration": True, "failure_func": lambda:None, "solve": lambda env, planner:solve_push_to_target_with_peg(env,planner,self.cube,self.goal_site,self.direction1,self.obj_flag), "segment":[self.cube,self.goal_site], }, { "func": lambda: static_check(self, timestep=int(self.elapsed_steps), static_steps=30), "name": "static", "subgoal_segment":f"static", "demonstration": True, "failure_func": None, "solve": lambda env, planner: [solve_hold_obj(env, planner, static_steps=30)], }, { "func": lambda:reset_check(self), "name": "NO RECORD", "subgoal_segment":f"NO RECORD", "demonstration": True, "failure_func": None, "specialflag":"reset pegs", "solve": lambda env, planner: [solve_strong_reset(env,planner)], }, { "func": lambda: is_any_obj_pickup_flag_currentpickup(self, objects=[self.grasp_target,self.grasp_target_false]), "name": "Pick up the peg", "subgoal_segment":f"Pick up the peg at <>", "choice_label": "pick up the peg", "demonstration": False, "failure_func": lambda:[ is_obj_pickup(self, obj=self.cube), is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2),], "solve": lambda env, planner:grasp_and_lift_peg_side(env, planner, env.grasp_target), "segment":self.grasp_target }, { "func": lambda: is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2,must_gripper_open=True), "name": f"Hook the cube to the target with the peg", "subgoal_segment":f"Hook the cube at <> to the target at <> with the peg", "choice_label": "hook the cube to the target with the peg", "demonstration": False, "failure_func": lambda:None, "solve": lambda env, planner:solve_push_to_target_with_peg(env,planner,self.cube,self.goal_site,self.direction2,self.obj_flag), "segment":[self.cube,self.goal_site], }, ] #test using gripper/grasp =false if self.way=="gripper_push": tasks = [{ "func": lambda: is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2,must_gripper_open=True), "name": "Close the gripper and push the cube to the target", "subgoal_segment":f"Close the gripper and push the cube at <> to the target at <>", "choice_label": "close gripper and push the cube to the target", "demonstration": True, "failure_func": lambda: [is_obj_pickup(self, obj=self.cube), is_obj_pickup(self, obj=self.grasp_target), is_obj_pickup(self, obj=self.grasp_target_false)], "solve": lambda env, planner:solve_push_to_target(env,planner,self.cube,self.goal_site), "segment":[self.cube,self.goal_site], }, { "func": lambda: static_check(self, timestep=int(self.elapsed_steps), static_steps=60), "name": "static", "subgoal_segment":f"static", "demonstration": True, "failure_func": None, "solve": lambda env, planner: [solve_hold_obj(env, planner, static_steps=60)], }, { "func": lambda:reset_check(self), "name": "NO RECORD", "subgoal_segment":f"NO RECORD", "demonstration": True, "failure_func": None, "specialflag":"reset pegs", "solve": lambda env, planner: [solve_strong_reset(env,planner)], }, { "func": lambda: is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2,must_gripper_open=True), "name": "Close the gripper and push the cube to the target", "subgoal_segment":f"Close the gripper and push the cube at <> to the target at <>", "choice_label": "close gripper and push the cube to the target", "demonstration": False, "failure_func": lambda: [is_obj_pickup(self, obj=self.cube), is_obj_pickup(self, obj=self.grasp_target), is_obj_pickup(self, obj=self.grasp_target_false)], "solve": lambda env, planner:solve_push_to_target(env,planner,self.cube,self.goal_site), "segment":[self.cube,self.goal_site], }, ] if self.way=="grasp_putdown": tasks = [ { "func": lambda: is_obj_pickup(self, obj=self.cube), "name": "Pick up the cube", "subgoal_segment":f"Pick up the cube at <>", "choice_label": "pick up the cube", "demonstration": True, "failure_func": lambda: [is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2), is_obj_pickup(self, obj=self.grasp_target), is_obj_pickup(self, obj=self.grasp_target_false)], "solve": lambda env, planner:[solve_pickup(env, planner, obj=self.cube),], "segment":[self.cube], }, { "func": (lambda: is_obj_dropped_onto(self,obj=self.cube,target=self.goal_site)), "name": "place the cube onto the target", "subgoal_segment":f"place the cube onto the target at <>", "choice_label": "place the cube onto the target", "demonstration": True, "failure_func": None, "solve": lambda env, planner: [solve_putonto_whenhold(env, planner,target=self.goal_site)], "segment":[self.goal_site], }, { "func": lambda: static_check(self, timestep=int(self.elapsed_steps), static_steps=60), "name": "static", "subgoal_segment":f"static", "demonstration": True, "failure_func": None, "solve": lambda env, planner: [solve_hold_obj(env, planner, static_steps=60)], }, { "func": lambda:reset_check(self), "name": "NO RECORD", "subgoal_segment":f"NO RECORD", "demonstration": True, "failure_func": None, "specialflag":"reset pegs", "solve": lambda env, planner: [solve_strong_reset(env,planner)], }, { "func": lambda: is_obj_pickup(self, obj=self.cube), "name": "Pick up the cube", "subgoal_segment":f"Pick up the cube at <>", "choice_label": "pick up the cube", "demonstration": False, "failure_func": lambda: [is_obj_pushed_onto(self,self.cube,self.goal_site,distance_threshold=self.cube_half_size*2*1.2), is_obj_pickup(self, obj=self.grasp_target), is_obj_pickup(self, obj=self.grasp_target_false)], "solve": lambda env, planner:[solve_pickup(env, planner, obj=self.cube),], "segment":[self.cube], }, { "func": (lambda: is_obj_dropped_onto(self,obj=self.cube,target=self.goal_site)), "name": "place the cube onto the target", "subgoal_segment":f"place the cube onto the target at <>", "choice_label": "place the cube onto the target", "demonstration": False, "failure_func": None, "solve": lambda env, planner: [solve_putonto_whenhold(env, planner,target=self.goal_site)], "segment":[self.goal_site], }, ] # Store task list for RecordWrapper use self.task_list = tasks # Use encapsulated sequence task check function if(self.use_demonstrationwrapper==False):# change subgoal after planner ends during recording if solve_complete_eval==True: allow_subgoal_change_this_timestep=True else: allow_subgoal_change_this_timestep=False else:# during demonstration, video needs to call evaluate(solve_complete_eval), video ends and flag changes in demonstrationwrapper if solve_complete_eval==True or self.demonstration_record_traj==False: allow_subgoal_change_this_timestep=True else: allow_subgoal_change_this_timestep=False #allow_subgoal_change_this_timestep=True all_tasks_completed, current_task_name, task_failed,_ = sequential_task_check(self, tasks,allow_subgoal_change_this_timestep=allow_subgoal_change_this_timestep) if task_failed: self.failureflag = torch.tensor([True]) logger.debug(f"Task failed: {current_task_name}") # If static_check succeeds or all tasks completed, set success flag if all_tasks_completed and not task_failed: self.successflag = torch.tensor([True]) return { "success": self.successflag, "fail": self.failureflag, } def compute_dense_reward(self, obs: Any, action: torch.Tensor, info: Dict): tcp_to_obj_dist = torch.linalg.norm( self.agent.tcp_pose.p - self.agent.tcp_pose.p, axis=1 ) reaching_reward = 1 - torch.tanh(5 * tcp_to_obj_dist) reward = reaching_reward*0 return reward def compute_normalized_dense_reward( self, obs: Any, action: torch.Tensor, info: Dict ): return self.compute_dense_reward(obs=obs, action=action, info=info) / 5 #Robomme def step(self, action: Union[None, np.ndarray, torch.Tensor, Dict]): obs, reward, terminated, truncated, info = super().step(action) timestep = int(info["elapsed_steps"]) if self.reset_in_proecess==True: for i, peg in enumerate(self.pegs): peg.set_pose(self.peg_init_poses_2[i]) if peg.dof > 0: zero = np.zeros(peg.dof) peg.set_qpos(zero) peg.set_qvel(zero) self.cube.set_pose(self.cube_init_pose_2) #self.goal_site_2.set_pose(sapien.Pose(p=self.goal_site_2_pose_p[0],q=self.goal_site_2_pose_q[0])) goal2_p = np.array(self.goal_site_2_pose_p, copy=True) goal2_q = np.array(self.goal_site_2_pose_q, copy=True) self.goal_site.set_pose(sapien.Pose(p=goal2_p[0],q=goal2_q[0])) #print("reset goal site to",goal2_p[0],goal2_q[0]) return obs, reward, terminated, truncated, info