Research On Navigation And Obstacle Avoidance Methods For Unmanned Vehicles On Unstructured Road

With the development of technologies such as artificial intelligence and deep learning,unmanned vehicles have emerged as a popular research direction in the field of automation.Due to their mobility,they can play a significant role in areas such as firefighting,agriculture,,and military operations.However,these scenarios often involve unstructured road environments with complex and dynamic navigable areas,making navigation and obstacle avoidance for unmanned vehicles on unstructured roads a challenging problem.This paper focuses on addressing this problem by analyzing,applying,and improving relevant techniques,ranging from unstructured road segmentation to map construction,and finally achieving navigation and obstacle avoidance for unmanned vehicles on unstructured roads.The specific contributions of this paper are summarized as follows:(1)In response to unstructured road scenarios,a dataset suitable for semantic segmentation on unstructured roads is selected and category redefined based on the Indian driving dataset.Preprocessing operations are performed to improve model training and enhance its generalization capability.This paper proposes a lightweight semantic segmentation network algorithm based on Bi Se Net for unstructured road scenes.It utilizes a lightweight backbone network and incorporates depth-wise separable convolutions to optimize speed control.Channel attention is introduced during feature fusion to adaptively select important features,suppress redundant information,and improve the accuracy of unstructured road segmentation.With only 1.11×10~6model parameters,the proposed algorithm achieves an 18.83%speed improvement compared to Bi Se Net and an F1-score of 96.74%.It demonstrates certain advantages over other mainstream semantic segmentation models and provides valuable insights for the safe operation of unmanned vehicles in unstructured road scenarios.(2)To meet the requirements of high-precision maps for unmanned vehicle navigation and obstacle avoidance,various forms of maps are compared,analyzing their strengths and limitations.The grid-based map is chosen for unmanned vehicle navigation and obstacle avoidance.Additionally,as using only Li DAR for building grid maps lacks semantic information,this paper proposes the utilization of an improved Bi Se Net semantic segmentation model to perform semantic segmentation on images captured by a stereo camera.The segmented images are then pixel-aligned with the depth map,resulting in a grid map that incorporates semantic information.(3)Based on the grid map,an analysis is conducted on the navigation algorithm based on global path planning and the obstacle avoidance algorithm based on local path planning.The strengths and limitations are summarized.Focusing on the issue of the traditional Dynamic Window Approach(DWA)not considering the initial pose direction and path node pose direction,a pose adjustment function is designed.To enhance the ability to avoid unknown obstacles,the evaluation function of DWA is improved by considering the impact of both known and unknown obstacles on planning,thus improving the safety of path planning.Simulation results demonstrate that the improved approach reduces planning time by 15.56%and shortens path length by 4.22%in static environments.In dynamic environments,planning time is reduced by 11.78%and path length is shortened by 8.48%.These results validate the effectiveness of the proposed improvements.(4)The unmanned vehicle’s hardware and software system is designed,and an analysis is conducted on its kinematic model to establish a foundation for unmanned vehicle navigation and obstacle avoidance in unstructured road scenarios.The hardware platform includes a tracked differential drive chassis,Li DAR,stereo camera,and other sensor modules,as well as a mini-PC and Jetson TX2 control center.The software platform is based on the Ubuntu 18.04 operating system,utilizing the Melodic version of the ROS system.Development is carried out using the navigation framework,incorporating A*algorithm and an improved Dynamic Window Approach for navigation and obstacle avoidance.Finally,experimental validation is performed using the built tracked unmanned vehicle,successfully achieving the implementation of navigation and obstacle avoidance functions in unstructured road scenarios.