Reward-Guided Domain Randomization (RGDR)
A project proposing an adaptive domain-randomization framework that improves robustness and generalization of autonomous-driving RL policies through reward-guided sampling.
View Code on GitHub
Selected Projects
Robust Reinforcement Learning for Autonomous Driving in Uncertain Environments
A history-dependent reinforcement-learning framework that improves policy adaptability and robustness in dynamic, partially observable driving environments using randomized dynamics, context-aware rewards, and LSTM-based policy optimization.