| Autonomous Obstacle avoidance system is one of the core systems for mobile robot.It is important to improve the safety and reliability of mobile robot in unstructured environment.The traversable condition in unstructured environment is relative complex with various traversable region types and blur boundaries.Traditional obstacle avoidance system depends on geometric information and low-level visual features,it lacks high-level semantic information.Therefore,it’s hard for traditional obstacle avoidance system to distinguish between various types of traversable area in unstructured environment,such as road,sidewalk and grass.Furthermore,traditional obstacle avoidance system does not measure the effect of the terrain and semantic factor for the system safety.These may cause unstable movement,even failure.In order to solve these problems,this thesis focus on the autonomous obstacle avoidance system in the unstructured environment,and the main research contents are the mapping generation method and path planning method.Firstly,a semantic traversable probabilistic map generation method is proposed,the map generation method takes RGB images as input,fuses visual semantic information in the grid map,then uses distance transformation and probability transformation to generate probabilistic traversable map.Based on the generated semantic probabilistic traversable map and elevation map using LIDAR pointcloud,an advanced dynamic window approach is also proposed.By adding terrain factor and semantic factor in the cost function,the advanced method improves the path planning result.At last,the entire obstacle avoidance system on the mobile robot platform is built in this thesis.Several tests are conducted on the platform.Experimental results show that our map generation method can better express the traversable area compare to the traditional elevation map.Furthermore,by adding the proposed path planning method,the safety and reliability of our path planning are increased. |
PDF Full Text Request