AI GitHub
ManiSkill for robot manipulation experiments
Train and evaluate manipulation policies in simulated tasks with visual and state observations.
Key Takeaways
- ManiSkill defines tasks and robot interfaces around simulated environments.
- Choose the simulator mode, task and robot configuration, then inspect sample observations.
- Choose one manipulation task and compare state-based and visual observations.
- Simulation settings and task versions shape reported success.
How it works
ManiSkill defines tasks and robot interfaces around simulated environments. Parallel simulation helps run many copies while policies learn from observations or demonstrations.
What you need
Choose the simulator mode, task and robot configuration, then inspect sample observations. Keep any demonstrations aligned to the action representation used by the policy.
What to try first
Choose one manipulation task and compare state-based and visual observations. Inspect success criteria and test a trained policy on changed starting positions.
Add a distractor object and shift the initial pose within the documented task variation. Inspect whether failures come from perception or control.
Where to look for failures
Simulation settings and task versions shape reported success. Evaluate transfer explicitly before moving from simulation to a real robot.
Related Franklin pages
Start with saved data or a simulator before deploying a policy on hardware. Verify observation layout, action conventions and task success using a known reference episode.
- Study manipulation tasks in simulated kitchens with varied scenes and objects. RoboCasa for simulated household manipulation
- Record robot demonstrations and train policies using reusable data and control tools. LeRobot for robot demonstrations and policy training
Franklin’s robotics coverage gives examples of where simulation, training and hardware meet. Franklin’s robotics coverage
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗