Asynchronous Reinforcement Learning for UR5 Robotic Arm

This is the implementation for asynchronous reinforcement learning for UR5 robotic arm. This repo consists of two parts, the vision-based UR5 environment, which is based on the SenseAct framework, and a asynchronous learning architecture for Soft-Actor-Critic. (Our implementation of SAC is partly borrowed from here)

Trained results:

Reaching	Tracking

Installation:

If you have the physical setup shown above (an UR5 robotic arm, an USB camera, and a monitor which are all connected to a Linux workstaion with GPU support), follow the steps below to install

1. install SenseAct framework
2. git clone https://github.com/YufengYuan/ur5_async_rl
3. cd ur5_async_rl
4. pip3 install -r requirements.txt

Instructions

To train an agent with asynchronous learning updates on the our reaching task, run:

python3 ur5_train.py \
      --target_type reaching \
      --seed 0 \
      --batch_size 128 \ 
      --episode_length 4 \ 
      --dt 0.04 \ 
      --image_width 160 \ 
      --image_height 90 \ 
      --init_step 100 \
      --async

where --async is the flag to enable asynchronous learning updates, without it, all computations will be processed sequentially. You may also need to change UR5 ip and camera_id in env/ur5_wrapper.py so that the system can properly detect the hardware.

The console output is available in a form:

| train | E: 1 | S: 1000 | D: 0.8 s | R: 0.0000 | BR: 0.0000 | ALOSS: 0.0000 | CLOSS: 0.0000 | NUM: 0.0000

a training entry decodes as:

E - total number of episodes 
S - total number of environment steps
D - duration in seconds of 1 episode
R - episode reward
BR - average reward of sampled batch
ALOSS - average loss of actor
CLOSS - average loss of critic
NUM - number of gradient updates performed so far