Implementation:Haosulab ManiSkill TwoRobotStackCube

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

Overview

Concrete implementation of the two-robot cooperative cube stacking task environment in ManiSkill.

Description

The TwoRobotStackCube environment features two Panda robot arms that must cooperate to stack two cubes. The green cube is near the right robot and the blue cube is near the left robot. One robot must place the green cube on a target region, then the other must stack the blue cube on top. This requires coordination and sequential planning between the two robots.

Registered as TwoRobotStackCube-v1 with max_episode_steps=100. Uses MultiAgent with two Panda robots. Reward modes include "normalized_dense", "dense", "sparse", and "none".

Randomizations: both cubes have random z-axis rotations, positions are set such that each robot can only reach one cube initially. A target region (marked by a red/white ground target) is randomized for the base cube placement. Success requires the blue cube to be stacked on top of the green cube on the target.

Usage

Use this environment for multi-agent cooperative manipulation research involving sequential, multi-step coordination between two robot arms.

Code Reference

Source Location

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

Signature

@register_env("TwoRobotStackCube-v1", max_episode_steps=100)
class TwoRobotStackCube(BaseEnv):
    agent: MultiAgent

Import

import gymnasium as gym
import mani_skill.envs
env = gym.make("TwoRobotStackCube-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 two Panda robots

Outputs

Name	Type	Description
obs	dict/array	Observation including both robots' states, cube poses, target position
reward	float	Dense reward based on sequential stacking milestones
terminated	bool	Whether episode ended by success
truncated	bool	Whether episode hit max steps (100)
info	dict	Contains success flag and stage progress

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("TwoRobotStackCube-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