Before you start with this challenge please FORK this repo and implement your solution in the forked repo.
According to DJI, the Robomaster robot needs to be able to navigate in the battle arena autonomously and fire on target if opponent is spotted. Therefore, we believe Reinforcement Learning is best suitted for this task. Thus, we are recruiting students who are good at designing Reinforcement Learning Models.
Your task, should you choose to accept it, is to create an agent throught Reinforcement Learning. Design an agent able to navigate in this Pygame arena autonomously and preferably eliminate the opponent(s) at the end of this simulation.
Please define your agent in the rl_agent_actor.py: rl_agent_actor.py
All the required libraries are found in requirements.txt
To install required libraries:
pip install -r requirements.txt
You are free to modify any files in this repository, including the environment variables (HP of Robot, Movement Speed, etc). However, your submission will be tested with the default environment against the default, unmodified rule-based actor in actor.py.
This simulator is adapted from UBC Robomaster AI Challenge Simulator
RoboMaster AI Challenge Simulator (RMAICS), is a 2D simulation environment designed for the ICRA 2021 RoboMaster AI Challenge. Its main function is to provide a simulation environment for intelligent decision-making groups to train neural networks.
The simulation consists of two levels:
The high-level training interface class:
rmaics
The low-level implementation class:kernel
See the following manuals for further information.
- High-level training interface in
rmaics.py: rmaics.md - Low-level implementation in
kernel.py: kernel.md - Instructions for
record_player.py: record_player.md - Instructions for controls: operation.md
- Parameter format: params.md
- Further improvements to
kernel.py: future.md