Implementation:Haosulab ManiSkill RotateSingleObjectInHand

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

Overview

Concrete implementation of the in-hand object rotation dexterous manipulation task in ManiSkill using the Allegro Hand robot.

Description

The RotateSingleObjectInHand environment requires the Allegro Hand Right (with touch sensors) to rotate an object in-hand to match a target orientation. The task has four difficulty levels (0-3):

• Level 0: Rotate a cube by a random angle around the z-axis only.
• Level 1: Rotate a cube to a random orientation.
• Level 2: Rotate a random YCB object by a random angle around the z-axis.
• Level 3: Rotate a random YCB object to a random orientation.

Registered variants include:

• RotateSingleObjectInHandLevel0-v1 (max_episode_steps=300)
• RotateSingleObjectInHandLevel1-v1 (max_episode_steps=300)
• RotateSingleObjectInHandLevel2-v1 (max_episode_steps=300, asset_download_ids=["ycb"])
• RotateSingleObjectInHandLevel3-v1 (max_episode_steps=300, asset_download_ids=["ycb"])

The supported robot is allegro_hand_right_touch. Reward modes include "dense", "normalized_dense", "sparse", and "none".

Usage

Use this environment for dexterous manipulation research, specifically in-hand object rotation. Higher difficulty levels introduce geometric variability through YCB objects and full 3D rotation targets.

Code Reference

Source Location

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

Signature

class RotateSingleObjectInHand(BaseEnv):
    agent: Union[AllegroHandRightTouch]
    def __init__(self, *args, robot_init_qpos_noise=0.02, obj_init_pos_noise=0.02,
                 difficulty_level: int = -1, num_envs=1, reconfiguration_freq=None, **kwargs): ...

@register_env("RotateSingleObjectInHandLevel0-v1", max_episode_steps=300)
class RotateSingleObjectInHandLevel0(RotateSingleObjectInHand): ...

Import

import gymnasium as gym
import mani_skill.envs
env = gym.make("RotateSingleObjectInHandLevel0-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 Allegro Hand

Outputs

Name	Type	Description
obs	dict/array	Observation including hand joint state, object pose, goal orientation
reward	float	Reward based on orientation distance to target
terminated	bool	Whether episode ended (object dropped)
truncated	bool	Whether episode hit max steps (300)
info	dict	Contains success, object_dropped flags

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("RotateSingleObjectInHandLevel0-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