Implementation:Haosulab ManiSkill MP Solutions Pattern

Field	Value
Implementation Name	MP Solutions Pattern
Type	Pattern Doc
Domain	Motion_Planning
Source File	`mani_skill/examples/motionplanning/panda/run.py` (L13-76)
Date	2026-02-15
Repository	Haosulab/ManiSkill

Overview

The MP Solutions Pattern implements the solver-selection registry for the motion planning demo generation pipeline. It defines the MP_SOLUTIONS dictionary that maps environment IDs to their corresponding solver functions, and provides the command-line interface and main execution loop for generating trajectories.

Description

The pattern consists of three components:

Registry dictionary (MP_SOLUTIONS): A Python dict mapping string environment IDs to callable solver functions imported from mani_skill.examples.motionplanning.panda.solutions.

Argument parser (parse_args): Defines all CLI parameters for controlling trajectory generation.

Main execution loop (_main): Creates the environment, wraps it with RecordEpisode, and iterates through trajectory generation, calling the selected solver for each episode.

Usage

# Generate 10 PickCube trajectories with video recording
python -m mani_skill.examples.motionplanning.panda.run -e PickCube-v1 -n 10 --save-video

# Generate 50 successful StackCube trajectories using 4 processes
python -m mani_skill.examples.motionplanning.panda.run -e StackCube-v1 -n 50 --only-count-success --num-procs 4

# Generate with ray-traced rendering
python -m mani_skill.examples.motionplanning.panda.run -e PegInsertionSide-v1 -n 20 --shader rt

Code Reference

MP_SOLUTIONS Registry

from mani_skill.examples.motionplanning.panda.solutions import (
    solvePushCube, solvePickCube, solveStackCube, solvePegInsertionSide,
    solvePlugCharger, solvePullCubeTool, solveLiftPegUpright, solvePullCube,
    solveDrawTriangle, solveDrawSVG, solvePlaceSphere, solveStackPyramid,
)

MP_SOLUTIONS = {
    "DrawTriangle-v1": solveDrawTriangle,
    "PickCube-v1": solvePickCube,
    "StackCube-v1": solveStackCube,
    "PegInsertionSide-v1": solvePegInsertionSide,
    "PlugCharger-v1": solvePlugCharger,
    "PlaceSphere-v1": solvePlaceSphere,
    "PushCube-v1": solvePushCube,
    "PullCubeTool-v1": solvePullCubeTool,
    "LiftPegUpright-v1": solveLiftPegUpright,
    "PullCube-v1": solvePullCube,
    "DrawSVG-v1": solveDrawSVG,
    "StackPyramid-v1": solveStackPyramid,
}

CLI Parameters

Parameter	Type	Default	Description
`-e` / `--env-id`	str	`PickCube-v1`	Environment to solve. Must be a key in `MP_SOLUTIONS`.
`-n` / `--num-traj`	int	`10`	Number of trajectories to generate.
`--only-count-success`	flag	`False`	Only save and count successful trajectories.
`-o` / `--obs-mode`	str	`none`	Observation mode (observations can be replayed later).
`-b` / `--sim-backend`	str	`auto`	Simulation backend: `auto`, `cpu`, or `gpu`.
`--render-mode`	str	`rgb_array`	Render mode: `sensors` or `rgb_array`.
`--record-dir`	str	`demos`	Directory for recorded trajectory output.
`--num-procs`	int	`1`	Number of parallel CPU processes.
`--save-video`	flag	`False`	Whether to save video files.
`--vis`	flag	`False`	Open a GUI to visualize the solution live.
`--shader`	str	`default`	Shader for rendering: `default`, `rt`, or `rt-fast`.
`--traj-name`	str	(timestamp)	Name for the trajectory `.h5` file.

Execution Loop (L44-125)

def _main(args, proc_id: int = 0, start_seed: int = 0) -> str:
    env_id = args.env_id
    env = gym.make(
        env_id,
        obs_mode=args.obs_mode,
        control_mode="pd_joint_pos",
        render_mode=args.render_mode,
        sensor_configs=dict(shader_pack=args.shader),
        human_render_camera_configs=dict(shader_pack=args.shader),
        viewer_camera_configs=dict(shader_pack=args.shader),
        sim_backend=args.sim_backend,
    )
    if env_id not in MP_SOLUTIONS:
        raise RuntimeError(
            f"No already written motion planning solutions for {env_id}. "
            f"Available options are {list(MP_SOLUTIONS.keys())}"
        )
    # ... wraps env with RecordEpisode, then loops calling solve(env, seed=seed) ...

I/O Contract

Direction	Data	Format
Input	Environment ID	String key in `MP_SOLUTIONS`
Input	CLI arguments	Parsed via `argparse`
Output	Trajectory file	`{record_dir}/{env_id}/motionplanning/{traj_name}.h5`
Output	Metadata file	`{record_dir}/{env_id}/motionplanning/{traj_name}.json`
Output	Video files (optional)	`{record_dir}/{env_id}/motionplanning/{video_id}.mp4`
Return	H5 file path	`str` (returned by `_main`)

Usage Examples

# Programmatic usage (not typical; usually invoked via CLI)
from mani_skill.examples.motionplanning.panda.run import _main, parse_args
import sys

sys.argv = ["run.py", "-e", "PickCube-v1", "-n", "5", "--record-dir", "my_demos"]
args = parse_args()
output_path = _main(args, proc_id=0, start_seed=0)
print(f"Trajectories saved to: {output_path}")

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment