AI Agents for Supply Chain: Path Following Control Using Adaptive Kalman Filter Reinforcement Learning

Path following control is a fundamental problem in robotics and autonomous systems, essential for the navigation of vehicles, drones, and mobile robots. The goal is to maintain a system on a predefined path despite external perturbations and complex system dynamics. Reinforcement learning (RL) offers a promising approach to overcome these limitations, allowing agents to learn control rules for following the optimal road route correlated with environmental conditions. Among RL algorithms, the Soft Actor-Critic (SAC) method stands out by combining the advantages of calculations made based on variables and rules. They will use an “actor” to select actions and a “critic” to evaluate the quality of these actions, facilitating efficient and stable learning. This paper proposes the integration of a Parallel Adaptive Kalman Filters (KF) with a Soft Actor-Critic (SAC) agent to create a highly robust system for path following control (PFC). This architecture combines optimal state estimation with maximum-entropy reinforcement learning to handle sensor noise, environmental disturbances, and continuous actuator controls.