Reinforcement Learning for Decision Making

[Bill-Haoyu-Lin]

Description:

For Full-autonomous robot and semi-autonomous robots, we will need reinforcement learning to give it certain degree of intelligence on the battleground to do decision making.

linked issues

27 simulation for RMUL and RMUC

24 State Machine

Tasks

• [ ] 3v3 simulation
• [ ] 3v3 RL model
• [ ] 7v7 simulation
• [ ] 7v7 RL model

[Allen-ENGI]

Testing Repo: https://github.com/LoveThinkinghard/RoboMaster-AI-Challenge-Simulator-2D Important document to modify and for understanding: params.md I can translate the files into English if needed.

As mentioned in the original repo, the modification should happen in the rmaics.py. The demo.py is just for familiarity with the env, the corresponding operations can be found in operation.md

Current goal: try to train one agent to move in the field without hitting the barriers. Issue: try to make the agent do different actions to reduce the hitting times. -> Maybe try the Q-learning method or implement a similar sample from gymnasium: https://gymnasium.farama.org/environments/box2d/car_racing/