This project aims at developing a self balancing control system for a two-wheeled robot using Reinforcement Learning. The plan is to train a model in an emulator with simulated data, and transfer the trained model to a real-life two wheeled robot and continue training with real-life data.
- Random search
- SARSA (G. A. Rummery , M. Niranjan)
- Q-learning (C. J. C. H. Watkins)
- Emulator: contains the code to train agent in simulated environment
- Models: contains a collection of agent models as listed above
- modify emulator environment to become more realistic
- re-train with the new environment
- deploy trained model in real robot