alexanderdurr/rl_rlbench

Reinforcement Learning with RLBench and PARL for Robot Arm

Implementation of Reinforcement Learning for Robot Arm

The implementation of reinforcement learning Algorithm TD3 for robot arm in RLBench Environment.

Current implementation is on reach target task, more tasks will be added in the future.

RLBench Environment	RLBench Gym Environment

Dependencies

• Python 3.6
• Paddlepaddle >= 1.8.2
• PARL >= 1.3.1
• RLBench
• Pyrep
• CoppeliaSim (previously called V-REP)

Install

RLBench is built around PyRep and CoppeliaSim. First head to the Pyrep github page and install them.

After that, go to RLBench github repo and install it .

Run

• To start the training program with gym GUI

$ pip3 install -r requirements.txt
$ python train_gym.py --train --mode='human'

Render mode: human, rgb_array and None

To test agent

$ python train_gym.py --test --mode='human'

• To start the training program with RLBench GUI

$ python train_rlbench.py --train
$ python train_rlbench.py --test

You can also run it headless with argument --headless

TO DO

• RLBench with Gym Env
• RLBench with Simple Task