CN-122018518-A - Obstacle avoidance method, device, equipment and medium based on laser radar and vector field histogram direction search

CN122018518ACN 122018518 ACN122018518 ACN 122018518ACN-122018518-A

Abstract

The application discloses an obstacle avoidance method, device, equipment and medium based on laser radar and vector field histogram direction search, which comprise the steps of simulating unmanned aerial vehicle interaction with environment based on a strategy network to collect track sequences, carrying out vector field histogram direction search on laser radar perception data of each time step to obtain a local obstacle avoidance direction of each time step, determining state value of each time step through a value network based on the local obstacle avoidance direction and state data of each time step, updating the strategy network and the value network by utilizing a near-end strategy optimization algorithm based on a data tuple and the state value of each time step to obtain a target strategy network, and controlling the unmanned aerial vehicle to deploy the target strategy network to avoid obstacle flight based on the target strategy network. The application adds the laser radar sensing data to the state data, and improves the obstacle avoidance flight effect by utilizing the memory capacity of the long and short memory neural network. Meanwhile, the local obstacle avoidance direction obtained by searching the vector field histogram direction of the laser radar sensing data is added into the state data input into the value network, so that an intelligent agent is guided to quickly learn the flight strategy of attaching the vector field histogram safety direction, and the obstacle avoidance efficiency and stability are greatly improved.

Inventors

WANG GANG
WANG JIKAI
ZHAO XINWEI
XIAO WEI
ZHANG XINHONG
WANG RUNQING
SUN JIAN
DENG FANG
CHEN JIE

Assignees

北京理工大学

Dates

Publication Date: 20260512
Application Date: 20251024

Claims (10)

1. The obstacle avoidance method based on the laser radar and the vector field histogram direction search is characterized by comprising the following steps of: Simulating interaction of the unmanned aerial vehicle with the environment based on a strategy network to collect a track sequence, wherein a data tuple of each time step in the track sequence comprises state data, actions, rewards and next state data, the state data comprises characteristic data of laser radar sensing data, unmanned aerial vehicle state information and target position information, and the strategy network comprises a long and short memory neural network and a multi-layer sensing machine; Carrying out vector field histogram direction search on the laser radar sensing data of each time step to obtain a local obstacle avoidance direction of each time step; Determining the state value of each time step through a value network based on the local obstacle avoidance direction and the state data of each time step; based on the data tuple and the state value of each time step, updating the strategy network and the value network by using a near-end strategy optimization algorithm to obtain a target strategy network; and controlling the unmanned aerial vehicle to deploy the target strategy network so as to carry out obstacle avoidance flight based on the target strategy network.
2. The obstacle avoidance method based on laser radar and vector field histogram direction search of claim 1, wherein the acquiring of the state data specifically comprises: Acquiring laser radar sensing data of an obstacle scene, and performing feature extraction on the laser radar sensing data to obtain feature data, wherein the laser radar sensing data is used for reflecting obstacle information in the obstacle scene; And splicing the characteristic data, the unmanned aerial vehicle state information and the target position information to obtain state data.
3. The obstacle avoidance method based on lidar and vector field histogram direction search of claim 1 or 2, wherein the drone state data comprises drone position information, drone pose information, drone speed information, drone orientation information, and drone thrust information.
4. The obstacle avoidance method based on lidar and vector field histogram direction search of claim 2, wherein the obtaining lidar perception data of the obstacle scene specifically comprises: obtaining obstacle distance distribution data under polar coordinates through a laser radar; And determining the nearest point cloud distance in each sector area by adopting a sector space division method with fixed dimension so as to obtain the laser radar perception data of the obstacle scene.
5. The obstacle avoidance method based on the search of the histogram directions of the lidar and the vector field according to claim 1, wherein the searching of the histogram directions of the vector field for the lidar sensing data of each time step to obtain the local obstacle avoidance direction of each time step specifically comprises: Searching an optimized direction closest to the target direction and smaller than a preset distance threshold value in the laser radar sensing data based on the vector field histogram direction for each time step; if the optimized direction is searched, the optimized direction is used as a local obstacle avoidance direction of the time step; and if the optimal direction is not searched, taking the local obstacle avoidance direction of the previous time step of the time step as the local obstacle avoidance direction of the time step.
6. The method of claim 1, wherein the rewards are determined based on pre-constructed rewards functions including a high-speed distance rewards function for shortening the time for the drone to reach a target location, a speed direction rewards function for exciting the drone to fly at high speed and toward the target direction, a safety rewards function for improving obstacle avoidance safety, a normal vector rewards function for limiting the drone normal vector tilt angle, a normal vector constraint rewards function for prolonging the survival time of the drone in the environment, and a survival rewards function.
7. The obstacle avoidance method based on lidar and vector field histogram direction search of claim 1 wherein, after updating the policy network and the value network with a near-end policy optimization algorithm based on the data tuples and state values for each time step to obtain a target policy network, the method further comprises: Exporting the target policy network; constructing an unmanned aerial vehicle obstacle flight simulation scene in a simulation environment; and determining a flight control instruction of the simulation unmanned aerial vehicle in the unmanned aerial vehicle obstacle flight simulation scene based on the target strategy so as to realize migration verification of the target strategy network from simulation to simulation, and deploying the verified target strategy network to the unmanned aerial vehicle.
8. The obstacle avoidance method based on lidar and vector field histogram direction search of claim 1 wherein, in updating the strategic network and the value network with a near-end strategic optimization algorithm, a dominance function is determined by generalized dominance estimation to constrain the updating of the strategic network and a mean square error loss function is used to optimize the value network.
9. The obstacle avoidance device based on the laser radar and the vector field histogram direction search is characterized by comprising the following specific components: The system comprises a collection module, a control module and a control module, wherein the collection module is used for simulating interaction of the unmanned aerial vehicle with the environment based on a strategy network to collect a track sequence, wherein a data tuple of each time step in the track sequence comprises state data, actions, rewards and next state data, the state data comprises characteristic data of laser radar sensing data and unmanned aerial vehicle state information, and the strategy network comprises a long and short memory neural network and a multi-layer sensing machine; The training module is used for searching the vector field histogram direction of the laser radar sensing data of each time step to obtain the local obstacle avoidance direction of each time step, determining the state value of each time step through the value network based on the local obstacle avoidance direction and the state data of each time step, and updating the strategy network and the value network by using a near-end strategy optimization algorithm based on the data tuple and the state value of each time step to obtain a target strategy network; And the control module is used for controlling the unmanned aerial vehicle to deploy the target strategy network so as to carry out obstacle avoidance flight based on the target strategy network.
10. A terminal device is characterized by comprising a processor and a memory; the memory has stored thereon a computer readable program executable by the processor; The steps in the obstacle avoidance method based on laser radar and vector field histogram direction search according to any one of claims 1-8 are implemented when the processor executes the computer readable program.

Description

Obstacle avoidance method, device, equipment and medium based on laser radar and vector field histogram direction search Technical Field The application relates to the technical field of unmanned aerial vehicle control, in particular to an obstacle avoidance method, device, equipment and medium based on laser radar and vector field histogram direction search. Background The unmanned aerial vehicle autonomous obstacle avoidance method generally divides an unmanned aerial vehicle obstacle avoidance task into three links, namely a sensing link, a planning link and a control link, wherein the sensing link is used for acquiring external obstacle information by adopting a sensor (such as a depth camera, a laser radar and other modules), the planning link is used for planning an obstacle avoidance track suitable for stable flight of the unmanned aerial vehicle by utilizing the external obstacle information, and the control link is used for tracking the obstacle avoidance track by adopting PID control, model prediction control and the like. However, when the planning link plans the obstacle avoidance track suitable for the stable flight of the unmanned aerial vehicle, a large amount of real-time calculation is needed to obtain the obstacle avoidance track, and the control link can not accurately track the obstacle avoidance track due to air resistance, planning model errors and other reasons, so that the autonomous obstacle avoidance effect of the unmanned aerial vehicle is affected. There is thus a need for improvements and improvements in the art. Disclosure of Invention The application aims to solve the technical problem of providing an obstacle avoidance method, device, equipment and medium based on laser radar and vector field histogram direction search aiming at the defects of the prior art. In order to solve the technical problems, the first aspect of the present application provides an obstacle avoidance method based on a laser radar and a vector field histogram direction search, wherein the obstacle avoidance method based on the laser radar and the vector field histogram direction search specifically includes: Simulating interaction of the unmanned aerial vehicle with the environment based on a strategy network to collect a track sequence, wherein a data tuple of each time step in the track sequence comprises state data, actions, rewards and next state data, the state data comprises characteristic data of laser radar sensing data, unmanned aerial vehicle state information and target position information, and the strategy network comprises a long and short memory neural network and a multi-layer sensing machine; Carrying out vector field histogram direction search on the laser radar sensing data of each time step to obtain a local obstacle avoidance direction of each time step; Determining the state value of each time step through a value network based on the local obstacle avoidance direction and the state data of each time step; based on the data tuple and the state value of each time step, updating the strategy network and the value network by using a near-end strategy optimization algorithm to obtain a target strategy network; and controlling the unmanned aerial vehicle to deploy the target strategy network so as to carry out obstacle avoidance flight based on the target strategy network. The obstacle avoidance method based on the laser radar and vector field histogram direction search, wherein the acquiring process of the state data specifically comprises the following steps: Acquiring laser radar sensing data of an obstacle scene, and performing feature extraction on the laser radar sensing data to obtain feature data, wherein the laser radar sensing data is used for reflecting obstacle information in the obstacle scene; And splicing the characteristic data, the unmanned aerial vehicle state information and the target position information to obtain state data. The unmanned aerial vehicle state data comprises unmanned aerial vehicle position information, unmanned aerial vehicle posture information, unmanned aerial vehicle speed information, unmanned aerial vehicle orientation information and unmanned aerial vehicle thrust information. The obstacle avoidance method based on the laser radar and vector field histogram direction search, wherein the obtaining the laser radar perception data of the obstacle scene specifically comprises the following steps: obtaining obstacle distance distribution data under polar coordinates through a laser radar; And determining the nearest point cloud distance in each sector area by adopting a sector space division method with fixed dimension so as to obtain the laser radar perception data of the obstacle scene. The obstacle avoidance method based on the search of the laser radar and the vector field histogram direction, wherein the search of the vector field histogram direction is performed on the laser radar sensing data of each time step, and the obtaining of the local