Skip to content

mininiii/reach-ball-env

Repository files navigation

reach-ball-env

This repository contains a custom OpenAI Gym environment for a three-wheel car simulation using PyBullet. The environment is designed to train an agent to reach a soccer ball using the SAC algorithm in Stable Baselines3. The soccer ball and the car are located randomly in each episode.

The implementation of a three-wheel car simulation using ROS2 and Gazebo can be found at https://github.com/mininiii/Ros2_moving_car.git

Demo Video

rl_demo.mov

Train

To start training the agent, run:

python3 sac_agent.py

If you want to visualize the training process, add the --gui True option:

python3 sac_agent.py --gui True

Tensorboard

tensorboard --logdir=./logs
tensorboard_success_rate

Visualize Trained Model

To visualize the trained model, run:

python3 visualize.py

If you want to change the model path and the number of episodes, add the --model_path {your_path} --len 5 option:

python3 visualize.py --model_path example/path/model --len 3

About

This repository contains a custom OpenAI Gym environment for a three-wheel car simulation using PyBullet.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages