Skip to content

Tool-use Robotic Benchmark built with Drake Simulation

License

Notifications You must be signed in to change notification settings

liruiw/Fleet-Tools

Repository files navigation

Fleet-Tools

Arxiv Fleet-Merge Website Policy-Composition Website

This repo is the simulation environment benchmark built with Gym API and Drake simulations for Franka Panda. The repository features tool-use tasks with scripted experts.

⚙️ Installation

  1. pip install -r requirements

🚶 Starting Commands

  1. These commands will use the hand-scripted joint-space kpam planner to plan for demonstration trajectories
  2. Run bash experiments/running_scripts/frankahammer_tool_datacollection_test_rgb.sh. open local_host:6006
  3. Run bash experiments/running_scripts/**_datacollection.sh to generate the data. And export to the global python path such that the training repo fleet_diffusion can import it and run evaluation.

🚶 Simulation Teleop Demonstrations

  1. Run python -m core.run run_expert=True teleop=True teleop_type=keyboard expert=HumanExpert task=FrankaDrakeSpatulaEnv task.use_meshcat=True. open local_host:6006 to use keyboard wsad to teleop.
  2. If you want to try Oculus Quest >=2. Run: python -m core.run run_expert=True teleop=True teleop_type=vr expert=HumanExpert task=FrankaDrakeSpatulaEnv task.use_meshcat=True Hold the trigger for moving.

💾 File Structure

├── ...
├── Fleet-Tools
|   |── assets 			# object assets
|   |── core 			# source code
|   |   |── agent 		# replay buffer to save data offline
|   |   |── expert  	# kpam expert to generate demonstrations 
|   |   └── ...
|   |── env 			# environment code
|   |── scripts 		# data preprocessing 
|   |── experiments     # data generation scripts and task configs

└── ...

🎛️ Note

  1. Code Quality Level: Tired grad student.
  2. Have Questions: Open Github issues or send an email.

License

MIT

Acknowledgement

Citation

If you find Fleet-Tools useful in your research, please consider citing:

@inproceedings{wang2023fleet,
author    = {Lirui Wang, Kaiqing Zhang, Allan Zhou, Max Simchowitz, Russ Tedrake},
title     = {Fleet Policy Learning Via Weight Mering and An Application to Tool Use},
booktitle = {Arxiv},
year      = {2023}
}

About

Tool-use Robotic Benchmark built with Drake Simulation

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published