Implementation:Haosulab ManiSkill SceneManipulationEnv

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

Overview

Concrete implementation of a base scene manipulation environment in ManiSkill for interacting with complex pre-built scenes.

Description

The SceneManipulationEnv provides a base environment for manipulation tasks in complex, pre-built scenes such as ReplicaCAD or AI2-THOR. It places a robot (Panda or Fetch) into a scene built by a configurable scene builder class. This environment has no success/failure metrics or rewards by default -- it is primarily for exploration and visualization.

Registered as SceneManipulation-v1 with max_episode_steps=200. Supported robots are panda and fetch (default: fetch).

The environment accepts a scene_builder_cls parameter (default "ReplicaCAD") to select which scene type to load. It also accepts build_config_idxs and init_config_idxs for controlling which scene configurations to sample. A fixed_scene parameter controls whether the scene is reconfigured on each reset.

Usage

Use this environment as a foundation for custom manipulation tasks in realistic indoor scenes. It supports scene builders for ReplicaCAD and AI2-THOR environments.

Code Reference

Source Location

• Repository: Haosulab_ManiSkill
• File: mani_skill/envs/scenes/base_env.py

Signature

@register_env("SceneManipulation-v1", max_episode_steps=200)
class SceneManipulationEnv(BaseEnv):
    SUPPORTED_ROBOTS = ["panda", "fetch"]
    agent: Union[Panda, Fetch]
    def __init__(self, *args, robot_uids="fetch",
                 scene_builder_cls: Union[str, SceneBuilder] = "ReplicaCAD",
                 build_config_idxs=None, init_config_idxs=None,
                 num_envs=1, **kwargs): ...

Import

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

I/O Contract

Inputs

Name	Type	Required	Description
obs_mode	str	No	Observation mode
reward_mode	str	No	Reward mode
control_mode	str	No	Control mode for the robot
scene_builder_cls	str/SceneBuilder	No	Scene builder to use (default: "ReplicaCAD")

Outputs

Name	Type	Description
obs	dict/array	Observation based on obs_mode
reward	float	Task reward (none by default)
terminated	bool	Whether episode ended
truncated	bool	Whether episode hit max steps (200)
info	dict	Environment information

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("SceneManipulation-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