Self-Balancing-Robot with Reinforcement Learning

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.

Agent models implemented

1. Random search
2. SARSA (G. A. Rummery , M. Niranjan)
3. Q-learning (C. J. C. H. Watkins)

Directories

• Emulator: contains the code to train agent in simulated environment
  • Models: contains a collection of agent models as listed above

To do:

• modify emulator environment to become more realistic
• re-train with the new environment
• deploy trained model in real robot