Implementation:Haosulab ManiSkill FMBAssembly

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

Overview

Concrete implementation of the Functional Manipulation Benchmark (FMB) Assembly task in ManiSkill.

Description

The FMBAssembly1Env is a simulation version of the Multi-Object Multi-Stage Manipulation Assembly1 task from the Functional Manipulation Benchmark (Luo et al.). The goal is to assemble parts together using a reorienting fixture onto a red board. The scene includes a board, yellow peg, purple U-shape, blue U-shape, green bridge, and a reorienting fixture.

Registered as FMBAssembly1Easy-v1 with max_episode_steps=500. The supported robot is panda. Reward modes include "sparse" and "none".

The bridge piece position is randomized with noise of +/-0.025m. Success is determined by the bridge being within 0.005m of its goal position on the board. Objects are loaded from local GLB/PLY asset files.

Usage

Use this environment for multi-stage assembly manipulation research. It tests the ability to grasp, reorient, and precisely place parts in a specific order.

Code Reference

Source Location

• Repository: Haosulab_ManiSkill
• File: mani_skill/envs/tasks/fmb/fmb.py

Signature

@register_env("FMBAssembly1Easy-v1", max_episode_steps=500)
class FMBAssembly1Env(BaseEnv):
    SUPPORTED_REWARD_MODES = ["sparse", "none"]
    SUPPORTED_ROBOTS = ["panda"]
    agent: Panda

Import

import gymnasium as gym
import mani_skill.envs
env = gym.make("FMBAssembly1Easy-v1")

I/O Contract

Inputs

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

Outputs

Name	Type	Description
obs	dict/array	Observation including TCP pose, board position, bridge pose
reward	float	Sparse reward when bridge is placed correctly
terminated	bool	Whether episode ended by success/failure
truncated	bool	Whether episode hit max steps (500)
info	dict	Contains success flag

Usage Examples

Basic Usage

import gymnasium as gym
import mani_skill.envs

env = gym.make("FMBAssembly1Easy-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