Vector Field Augmented Reinforcement Learning for Adaptive Motion Planning of Mobile Robots

Efficient and adaptive motion planning is crucial for mobile robots operating in complex and dynamic environments. However, existing motion planning methods struggle with high computational costs, poor data efficiency, and real-time adaptability. Therefore, here we propose vector field augmented reinforcement learning (VF-RL), an adaptive motion planning framework that integrates RL with vector field guidance, enabling mobile robots to safely complete real-world tasks. First, VF-RL constructs a safety-aware reference path using a composite vector field, refining it with kinematic constraints to ensure reliable collision-avoidance guidance. Then, our proposed receding-horizon RL, informed by the vector field guidance and exponential barrier function, generates optimal motions that effectively handle uncertain dynamics, enforcing safety constraints and respecting actuator limitations in real-time. To improve the adaptability of VF-RL, a data-driven model with online updates is designed to capture the intricate dynamic characteristics in ever-changing environments. Extensive simulations involving quadrotors and autonomous ground vehicles, alongside real-world experiments using the Hongqi E-HS3 autonomous platform, demonstrate the superior performance of the proposed VF-RL framework in computational efficiency, cost minimization, and obstacle avoidance capabilities compared to traditional optimization-based and RL-based methods.