Vanya Awasthi

This paper presents a lightweight Deep Q-Network framework for CPU-efficient Othello agents, investigating how dense reward shaping and exploration strategies affect learning efficiency. The approach combines state-dependent dense rewards with shallow, search-limited dynamic programming during action selection and evaluates eight exploration strategies through self-play and random-opponent experiments. Results show improved training stability and convergence, with Randomized Value Functions (RVF) demonstrating particularly strong exploration performance under computational constraints.

This paper proposes a reinforcement learning framework for autonomous rocket landing that removes the need for handcrafted control laws. The method introduces a custom 3D physics simulation incorporating gravity, thrust, rotational dynamics, off-center exhaust torque, and fuel consumption, while an RL agent learns thrust and attitude control from continuous state feedback. The framework formulates landing as a control optimization problem using reward shaping to jointly improve landing precision and stability, demonstrating the potential of RL for autonomous guidance in nonlinear rocket dynamics.