Comparative Analysis of Learning based and Classical Motion Planning Methods

Mobile robot navigation in unknown environments requires both effective decision‑making and robust obstacle avoidance. This paper compares two prevailing machine learning models, namely Reinforcement learning (RL) and Self‑imitation learning (SIL) vis-à-vis robot’s Potential Field Obstacle Avoidance (APF) navigation system and observes robots efficiency in reaching the goal with each of these methods. Reinforcement learning (RL) provides a framework in which an agent learns to make decisions by trial and error to maximize cumulative rewards, yet sparse rewards and exploration challenges can hinder training efficiency. Self‑imitation learning (SIL) addresses these issues by encouraging the agent to reproduce its own high‑return trajectories . Classical control approaches like the artificial potential field (APF) method compute attractive forces toward goals and repulsive forces away from obstacles . The paper uses a standard RL agent using proximal policy optimization (PPO), an RL agent augmented with SIL, and an APF controller for obstacle avoidance in a simulated e‑puck robot environment. We describe the environment, learning algorithms and potential‑field implementation, laying the foundation for a quantitative comparison of these techniques.