Added custom task (PullCubeTool-v1) with motion planner + Motion Planner for LiftPegUpright and PullCube #641
Conversation
|
|
||
| qpos[:, -2:] = 0.04 | ||
| self.agent.robot.set_qpos(qpos) | ||
| self.agent.robot.set_pose(sapien.Pose([-0.615, 0, 0])) |
There was a problem hiding this comment.
why are we initializing the robot here? The scene builder handles that already
| tcp_pose=self.agent.tcp.pose.raw_pose, | ||
| cube_pose=self.cube.pose.raw_pose, | ||
| tool_pose=self.l_shape_tool.pose.raw_pose, | ||
| ) |
There was a problem hiding this comment.
Following the setup of other envs (e.g. push cube), can you ensure the ground truth info (tool pose, cube pose) are only provided if the observation is state based
There was a problem hiding this comment.
this is still not resolved. cube pose and tool pose are always included
| self, obs: Any, action: torch.Tensor, info: Dict | ||
| ): | ||
| max_reward = 13.0 # 10 + 1 + 1 + 1 | ||
| return self.compute_dense_reward(obs=obs, action=action, info=info) / max_reward |
There was a problem hiding this comment.
have you tested the reward function with PPO? If so can you add an example script to examples/baselines/ppo/examples.sh that works?
examples/baselines/ppo/ppo.py
Outdated
| envs = gym.make(args.env_id, num_envs=args.num_envs if not args.evaluate else 1, **env_kwargs) | ||
| eval_envs = gym.make(args.env_id, num_envs=args.num_eval_envs, reconfiguration_freq=args.reconfiguration_freq, **env_kwargs) | ||
| envs = gym.make(args.env_id, num_envs=args.num_envs if not args.evaluate else 1, **env_kwargs, reward_mode = "dense") | ||
| eval_envs = gym.make(args.env_id, num_envs=args.num_eval_envs, reward_mode = "dense", reconfiguration_freq=args.reconfiguration_freq, **env_kwargs) |
| @@ -20,7 +20,7 @@ | |||
| import sapien | |||
|
|
|||
|
|
|||
| @register_env("PullCubeTool-v1", max_episode_steps=100) | |||
| @register_env(uid="PullCubeTool-v1", max_episode_steps=100) | |||
There was a problem hiding this comment.
no need to add uid here. other code doesn't do that.
There was a problem hiding this comment.
make sure to remove all your own files here
added bug fixes
| tcp_pose=self.agent.tcp.pose.raw_pose, | ||
| cube_pose=self.cube.pose.raw_pose, | ||
| tool_pose=self.l_shape_tool.pose.raw_pose, | ||
| ) |
There was a problem hiding this comment.
this is still not resolved. cube pose and tool pose are always included
merging w main
Viswesh-N
changed the title
Viswesh-N
changed the title
No description provided.