Feeding robot

This is Wang Yutong's part2 project for ME5406 in Department of Mechanical Engineering, National University of Singapore

Prerequisites

see requirement.txt to find the appropriate version of the software package

Training models

run.py is the main file for training a PR2 robot in static human enviroment using PPO

run_with_human.py is the main file for training a PR2 robot in human cooperation enviroment using PPO

sac.py is the main file for training a PR2 robot in static human enviroment using SAC

sac_cop.py is the main file for training a PR2 robot in human cooperation enviroment using SAC

above file can be used as follows:

python run.py --lr 3.0e-4 --gamma 0.99

See arguments.py arguments_with_human.py for a full list of available arguments and hyperparameters.

Use trained models

Run enjoy_play.py to verify the performance of the trained model in static human enviroment

Run enjoy_play_with_human.py to verify the performance of the trained model in human cooperation enviroment

python enjoy_play.py

Result(too big so I delete them)

summaries\ppo_result is the tensorboard result of ppo in static human enviroment

summaries\sac_result is the tensorboard result of sac in static human enviroment

summaries\cop_sac_result is the tensorboard result of sac in human cooperation enviroment

summaries\cop_ppo_result is the tensorboard result of ppo in human cooperation enviroment

my_video\cooperation is the video of trained model in human cooperation enviroment

my_video\static_human is the video of trained model in static human enviroment

Reference

This library is derived from code baseline:https://github.com/openai/baselines and Assistive Gym:https://github.com/Healthcare-Robotics/assistive-gym