A dynamic environment path planning method for a coal mine underground inspection unmanned aerial vehicle fusing an LSTM

By predicting the pose of dynamic obstacles using an LSTM neural network and combining it with the MPC model and RRT* algorithm, the problems of collision and poor real-time performance of traditional path planning algorithms in the dynamic environment of underground coal mines are solved, achieving more efficient obstacle avoidance and safe planning.

CN122360460APending Publication Date: 2026-07-10XIAN UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
XIAN UNIV OF SCI & TECH
Filing Date
2026-04-02
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Traditional path planning algorithms cannot effectively handle dynamic obstacles in the dynamic environment of underground coal mines, resulting in problems such as collisions and poor real-time performance.

Method used

An LSTM neural network is used to predict the future pose of dynamic obstacles. Combined with the MPC model and RRT* algorithm, a composite cost function is constructed for path planning, which realizes quantitative modeling of prediction uncertainty and a robust trade-off between safety margin and path efficiency.

Benefits of technology

It improves the obstacle avoidance capability and real-time performance of UAVs in the dynamic environment of underground coal mines, reduces the risk of collision, adapts to complex environments, and enhances the stability and safety of path planning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122360460A_ABST
    Figure CN122360460A_ABST
Patent Text Reader

Abstract

This invention relates to the field of navigation and path planning technology for unmanned aerial vehicles (UAVs) in coal mines. It discloses a dynamic environment path planning method for UAVs used for underground coal mine inspections, integrating LSTM. The method constructs a state space, uses an LSTM neural network to predict the future pose of dynamic obstacles, and outputs the expected mean and covariance matrix. The covariance matrix is ​​transformed into a probabilistic safety semi-axis to generate a confidence ellipsoid for the dynamic obstacles. The spatiotemporal occupancy of the dynamic obstacles at each time step is obtained through the confidence ellipsoid. The RRT* algorithm is run to perform hard collision detection across the entire prediction time domain. A composite cost function integrating geometric length cost and risk cost is constructed. Risk cost is evaluated for geometrically feasible candidate edges. The parent node selection and rewiring operations of the RRT* algorithm are completed to generate the optimal feasible path. The optimal feasible path is executed, and the entire path planning is completed. This invention improves the UAV's ability to avoid dynamic obstacles underground.
Need to check novelty before this filing date? Find Prior Art