Evaluation of neural euclidean signed distance fields for distance-aware local path planning

Abstract

This paper presents a 3D local path planner based on Neural Euclidean Signed Distance Fields (ESDFs). There are approaches in the literature that make use of ESDFs as representation of the environment for planning, but not on the use of neural ESDF for distance-aware local path planning. This paper evaluates the proposed approach and analyses the quality of the paths generated by planners based on online Neural ESDF. The distance field is based on the HIO-SDF network, and the planner exploits the analytical properties of the ESDFs. The experimental validation shows promising results on the applications of HIO-SDF networks for 3D local path planning.