A modified artificial potential field approach for real-time robot path planning via adaptive gradient neural network

Artificial potential field (APF) approach is a classical path planning algorithm widely employed for its simplicity. However, in intricate dynamic environments, conventional APF approaches intrinsically suffer from local minima and oscillation issues. In this regard, this paper constructs a modified artificial potential field (MAPF) approach based on adaptive gradient neural network (AGNN) presented to generate safe and smooth real-time paths. Specifically, a linear gravitational field is tactfully designed to ensure global convergence, while avoiding singularity problems of traditional formulations. Furthermore, a piecewise force amplification mechanism is intrinsically established to create distinct safety zones around obstacles, thereby providing enhanced repulsive forces for proactive collision avoidance. Simultaneously, a tangential escape mechanism is constructed to break the force equilibrium, effectively resolving the local minima trap problem. More importantly, an AGNN model is developed to adaptively regulate the motion step based on the gradient magnitude, which effectively suppresses oscillations in narrow passages. Furthermore, theoretical guarantees of global asymptotic convergence and obstacle impenetrability are rigorously established. Finally, numerical simulations and physics-engine simulations on the Webots robot simulation platform validate the effectiveness of the MAPF approach in generating feasible and collision-free paths.