Self-Balancing Platform via Reinforcement Learning
Sim-to-real RL for a 2-DOF self-balancing platform, deployed on low-power hardware.
Trained a 2-DOF self-balancing platform in the MuJoCo physics simulator using policy optimization. The training and inference pipeline combined MuJoCo, Gym, PyTorch, and TensorFlow. Achieved successful sim-to-real transfer via TFLite model quantization for low-power hardware inference, outperforming model-based approaches for balancing an object on a wobbling surface.
