Implementation:Haosulab ManiSkill PullCube

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

Overview

Concrete implementation of the cube pulling task environment in ManiSkill.

Description

The PullCubeEnv requires a robot to pull a cube toward a target position. The target is marked by a red/white circular goal region positioned behind the cube (relative to the robot).

Registered as PullCube-v1 with max_episode_steps=50. Supported robots: panda and fetch. Reward modes include "normalized_dense", "dense", "sparse", and "none".

Randomizations: cube xy position in [-0.1, 0.1], goal position fixed at cube_xy - [0.1 + goal_radius, 0]. Success requires the cube to be within the goal region on the table.

Usage

Use this environment for non-prehensile pulling manipulation research. Unlike pushing tasks, pulling requires the robot to position itself behind the object and drag it toward the goal.

Code Reference

Source Location

• Repository: Haosulab_ManiSkill
• File: mani_skill/envs/tasks/tabletop/pull_cube.py

Signature

@register_env("PullCube-v1", max_episode_steps=50)
class PullCubeEnv(BaseEnv):
    SUPPORTED_ROBOTS = ["panda", "fetch"]
    agent: Union[Panda, Fetch]

Import

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

I/O Contract

Inputs

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

Outputs

Name	Type	Description
obs	dict/array	Observation including cube pose, goal position, TCP pose
reward	float	Dense reward based on reaching and pulling the cube toward goal
terminated	bool	Whether episode ended by success
truncated	bool	Whether episode hit max steps (50)
info	dict	Contains success flag

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("PullCube-v1", obs_mode="state", render_mode="rgb_array")
obs, info = env.reset()
for _ in range(50):
    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