Implementation:Haosulab ManiSkill PickCube

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

Overview

Concrete implementation of the cube picking task environment in ManiSkill -- the baseline tabletop manipulation benchmark.

Description

The PickCubeEnv is the baseline manipulation task for testing robot grasping and placement capabilities. The robot must grasp a red cube and move it to a target goal position marked by a green sphere.

Registered variants:

• PickCube-v1 (max_episode_steps=50): Default with Panda robot.
• PickCubeSO100-v1 (max_episode_steps=50): With SO100 robot.
• PickCubeWidowXAI-v1 (max_episode_steps=50): With WidowXAI robot.

Supported robots: panda, fetch, xarm6_robotiq, so100, widowxai. Reward modes include "normalized_dense", "dense", "sparse", and "none".

Randomizations: cube xy position in [-0.05, 0.05], cube z-rotation, and goal position xy in [-0.05, 0.05] with z in [0, 0.3]. Success requires the cube to be within goal_thresh (default 0.025m) of the goal and the robot to be static (qvel < 0.2). Robot-specific configurations are loaded from PICK_CUBE_CONFIGS.

Usage

Use this environment as the primary benchmark for validating tabletop manipulation capabilities. It supports multiple robot embodiments and provides a well-tuned dense reward function.

Code Reference

Source Location

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

Signature

@register_env("PickCube-v1", max_episode_steps=50)
class PickCubeEnv(BaseEnv):
    SUPPORTED_ROBOTS = ["panda", "fetch", "xarm6_robotiq", "so100", "widowxai"]
    agent: Union[Panda, Fetch, XArm6Robotiq, SO100, WidowXAI]
    goal_thresh = 0.025

Import

import gymnasium as gym
import mani_skill.envs
env = gym.make("PickCube-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 TCP pose, cube pose, goal position, is_grasped
reward	float	Dense reward: reaching + grasping + placing + static
terminated	bool	Whether episode ended by success
truncated	bool	Whether episode hit max steps (50)
info	dict	Contains success, is_obj_placed, is_robot_static, is_grasped

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("PickCube-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