panda-rl-envs

Real Panda/FR3 (inverse) reinforcement learning tasks/environments. Used experimentally with VPACE.

Installation

1. Install panda-polymetis (and polymetis, following instructions in panda-polymetis repo).
2. Activate polymetis conda env.
3. Install this repo (pip install -e .)

Usage

1. Launch robot controller
   1. (real) bash /path/to/panda-polymetis/launch/real_robot.bash
   2. (sim) bash /path/to/panda-polymetis/launch/sim_robot.bash
2. Launch gripper controller
   1. (real) launch_gripper.py gripper=robotiq_2f gripper.comport=/dev/ttyUSB0
   2. (sim) (None, just a fake client)
3. Use as a regular gym environment, but without gym.make:

from panda_rl_envs import *

env = SimPandaReach()
obs = env.reset()

for i in range(10):
    obs, rew, done, info = env.step(env.action_space.sample())

Keyboard Control

To get quick access to teaching (without stopping the controller), keyboard control per-joint or in task space, and keyboard gripper control, run

python -m panda_polymetis.tools.keyboard_interface

Citation

If you find this repository useful for your work, please consider citing VPACE.