Implementation:Haosulab ManiSkill RotateValveEnv

Knowledge Sources	Haosulab_ManiSkill
Domains	Robotics, Simulation, Dexterous_Manipulation
Last Updated	2026-02-15 08:00 GMT

Overview

Concrete implementation of the valve rotation dexterous manipulation task in ManiSkill using the DClaw robot.

Description

The RotateValveEnv requires a DClaw tri-finger robot to rotate a Robel valve by a specified amount. The task has five difficulty levels (0-4):

• Level 0: Rotate valve by a quarter turn (pi/2 radians).
• Levels 1-3: Rotate valve by a half turn (pi radians), with increasing randomization of valve configuration.
• Level 4: Rotate valve by a full turn (2*pi radians) with random valve configurations.

Registered variants:

• RotateValveLevel0-v1 through RotateValveLevel4-v1 (max_episode_steps=200 each)

The supported robot is dclaw. Reward modes include "dense", "normalized_dense", "sparse", and "none". The valve is built as an articulated object using build_robel_valve.

Usage

Use this environment for researching dexterous manipulation with tri-finger grippers. The difficulty levels provide a curriculum from simple quarter-turn rotation to full-turn rotation with randomized valve geometry.

Code Reference

Source Location

• Repository: Haosulab_ManiSkill
• File: mani_skill/envs/tasks/dexterity/rotate_valve.py

Signature

class RotateValveEnv(BaseEnv):
    agent: Union[DClaw]
    def __init__(self, *args, robot_init_qpos_noise=0.02, valve_init_pos_noise=0.02,
                 difficulty_level: int = -1, **kwargs): ...

@register_env("RotateValveLevel0-v1", max_episode_steps=200)
class RotateValveLevel0Env(RotateValveEnv): ...

Import

import gymnasium as gym
import mani_skill.envs
env = gym.make("RotateValveLevel0-v1")

I/O Contract

Inputs

Name	Type	Required	Description
obs_mode	str	No	Observation mode
reward_mode	str	No	Reward mode: "dense", "normalized_dense", "sparse", "none"
control_mode	str	No	Control mode for the DClaw robot

Outputs

Name	Type	Description
obs	dict/array	Observation including finger joint state, valve angle, target angle
reward	float	Reward based on angular distance to target rotation
terminated	bool	Whether episode ended by success/failure
truncated	bool	Whether episode hit max steps (200)
info	dict	Contains success flag and rotation progress

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("RotateValveLevel0-v1", obs_mode="state", render_mode="rgb_array")
obs, info = env.reset()
for _ in range(100):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()
env.close()

Related Pages

• Environment:Haosulab_ManiSkill_Python_SAPIEN_Core