Repo for Rapid Motor Adaptation for Robotic Manipulator Arms (link). It's still under construction.

Installation

1. Clone the repo:

git clone --recurse-submodules https://github.com/yichao-liang/rma4rma

2. Create a conda environment:

conda env create -f environment_copy.yml

Example code

1. Base policy training:

python main.py -n 50 -bs 5000 -rs 2000 \
            --randomized_training --ext_disturbance --obs_noise \
            -e PickSingleYCB-v1 

2. Adaptation training:

python train_sb.py -n 50 -bs 5000 -rs 2000 \
    --randomized_training --ext_disturbance --obs_noise --use_depth_adaptation\
    -e PickSingleYCB-v1 --adaptation_training \
    --log_name PPO-ps-bs5000-rs2000-kl0.05-neps10-cr0.2-lr_scdl0-cr_scdl1-ms50-incObsN0_1 \
    --ckpt_name best_model.zip

3. Evaluate:

python train_sb.py --eval \
    --randomized_training --ext_disturbance --obs_noise --use_depth_adaptation\
    -e PickSingleYCB-v1 --ckpt_name best_model.zip \
    --log_name PPO-ps-bs5000-rs2000-kl0.05-neps10-cr0.2-lr_scdl0-cr_scdl1-ms50-incObsN0_1_stage2_dep_1