Potential Neural Fields: A New Path Planning Approach for Swift and Smooth Paths

In recent years, there has been a notable shift in neural network methodologies with the emergence of "neural fields," fully connected networks capable of encoding continuous signals across arbitrary dimensions. This paper introduces a novel path planning approach by integrating neural fields into the encoding of potential fields for map representation. The method involves fitting a neural network to a low-quality potential field signal derived from the map, enabling real-time decision-making during path planning. The key innovation lies in the neural network’s ability to interpolate data, enhancing the fidelity of the potential field representation. By efficiently providing surrounding potential values based on the robot’s location, the approach optimizes decision-making for path planning. Simulation results have proven that the proposed method has outperformed the well-known A* path planning technique in terms of execution speed and path quality.

I. Introduction

Robotic path planning is a fundamental aspect of autonomous systems, involving the determination of an optimal route for a robot to navigate from its current position to a designated goal while avoiding obstacles and adhering to specified constraints [1], [2]. This complex process encompasses the generation of a feasible trajectory in a given environment, considering factors such as terrain complexity, dynamic obstacles, and the robot’s physical limitations. Various algorithms, from classical techniques such as Dijkstra’s algorithm and A* to advanced probabilistic approaches such as Rapidly-exploring Random Trees (RRT) and Probabilistic Roadmaps (PRM), are employed to address the challenges of path planning [3]–[5]. The goal is to devise strategies that ensure collision-free movement while optimizing the path in terms of efficiency, time, and resource utilization. Robotic path planning has diverse applications, including manufacturing, logistics, healthcare, and exploration, contributing significantly to the advancement of autonomous robotics.