A dynamic riemannian manifold artificial potential field obstacle avoidance method based on PPO reinforcement learning

By using PPO reinforcement learning and Riemannian manifold artificial potential field method, the problem of insufficient risk differentiation and adaptability of traditional potential field method in complex scenarios is solved, realizing accurate obstacle avoidance and adaptive adjustment in multi-target environment, which is suitable for the obstacle avoidance needs of autonomous driving on urban roads.

CN121697672BActive Publication Date: 2026-06-30HEFEI UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HEFEI UNIV OF TECH
Filing Date
2025-12-25
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Traditional artificial potential field methods struggle to distinguish the directional risks of pedestrians and vehicles in complex multi-objective scenarios. The repulsion coefficients need to be preset manually and cannot be adjusted in real time, leading to local minima or unreachable targets. Furthermore, existing RL+APF fusion solutions lack adaptability to complex scenarios such as urban roads.

Method used

A dynamic Riemannian manifold artificial potential field obstacle avoidance method based on PPO reinforcement learning is adopted. A high-dimensional spatiotemporal state vector is constructed by fusing multi-source data, and the deformation parameters of the Riemannian manifold potential field are output. Combined with gravity, repulsion and virtual escape force, it can achieve differentiated protection and adaptive adjustment for obstacle types.

Benefits of technology

It achieves accurate collision risk matching for pedestrians, vehicles, and static obstacles, breaks local minimum deadlock, improves the system's adaptability and generalization in dynamic multi-objective environments, and adapts to various complex scenarios on urban roads.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121697672B_ABST
    Figure CN121697672B_ABST
Patent Text Reader

Abstract

This invention relates to the field of autonomous driving technology and provides a dynamic Riemannian manifold artificial potential field obstacle avoidance method based on PPO reinforcement learning. It categorizes obstacles into three types—pedestrians, vehicles, and static obstacles—using a target detection algorithm to clearly define the risk distribution characteristics of each category. Secondly, it constructs an 11-dimensional high-dimensional spatiotemporal state space to achieve early risk prediction. The PPO reinforcement learning algorithm is introduced, taking the high-dimensional state as input and outputting 4-dimensional potential field deformation parameters for the three types of obstacles, autonomously learning differentiated continuous potential field morphologies. Based on Riemannian geometry, a covariance matrix is ​​constructed to generate an anisotropic repulsive force field without abrupt changes, and a virtual escape force is superimposed to break local minima. The total potential field force is obtained by synthesizing the attraction, total repulsion, and escape force, outputting steering angle and acceleration commands that conform to vehicle dynamics constraints. This strategy constructs a complete obstacle avoidance framework, effectively alleviating the technical limitations of traditional solutions.
Need to check novelty before this filing date? Find Prior Art