chenzhike110 / SSL-RL

code and materials for paper "Multi-Stage Decision-Making Skill Learning for Soccer Robot", IEEE RCAR 2021

Multi-Stage Decision-Making Skill Learning for Soccer Robot

This is a project using Deep Reinforcement Learning algorithm to train soccer robot in Robocup SSL with basic skills and tragedies.

• Algorithm

  Twin Delayed Deep Deterministic Policy Gradient.

  the paper is here Addressing function approximation error in actor-critic methods

• Simulation

  RoboCup small size league official simulator grSim

  the repository is here grSim

• Skills

  Go to ball skill

  rotating shoot skill

• Tactics

  run and shoot

Training instruction

• Go to ball skill

  train_findball.py

• Shoot skill

  train_shooting.py

• Switch reaction environment

  ssl-env is a completely kinematic environment for find ball skill

  my_env is a interface communicating with grSim

• Evaluation

  run_and_kick.py test it in grSim simulator

  real_car.py test skills in real world

• The environment use UDP communication protocol. The IP and interface would change according to your setting.