Real Robot Challenge 2022 Training Code (Team Name: decimalcurlew)

This is the training code (team name: decimalcurlew) for the real-robot stage of the Real Robot Challenge 2022.

Training environments and datasets can be found at rrc_2022_datasets.

For simulation-based evaluation, we reccomend downgrading gym to [gym==0.24.1] after installing the rrc_2022_datasets.

Networks are trained using Tensorflow 2.9.0 and Python 3.8.

Usage

Experiments on all environemnts can be run by running:

$ sh run.sh

Experiments on a single environment can be run by calling:

$ python main.py --env "trifinger-cube-push-real-expert-v0"
$ python main.py --env "trifinger-cube-push-real-mixed-v0"
$ python main.py --env "trifinger-cube-lift-real-expert-v0"
$ python main.py --env "trifinger-cube-lift-real-mixed-v0"

Hyper-parameters can be modified with different arguments to main.py.