Motion Planning Of The Hexapod Robots Based On The Terrain Modeling By Geometric And Physical Characteristics

By imitating the mobility and flexibility of creatures,the legged robots can traverse complex environments by selecting discrete footholds,which traditional robots can not.Although typical legged robots currently have good motion control capabilities,they still lack the ability to explore unfamiliar environments autonomously.Autonomous behaviour ability has become the main factor restricting their practical application,and the realization of fully independent movements also requires intelligent environmental perception and decision capabi lities.The wild environment is exceptionally complicated,and the difference from urban roads is that the wild terrain is mostly soft and slippery,and there are geometric and non-geometric obstacles.It is hard to meet the actual needs through traditiona l geometric modeling.Therefore,it is necessary to propose a more comprehensive environmental model perception method and solve the motion planning problem under multiple constraints.Based on the aforementioned,this paper mainly focuses on multi-modal environmental perception and uses geometric information and physical information to make intelligent decision planning of hexapod robots.In terms of environmental modelling,this paper first used visual methods to construct environmental geometry and semantic models,constructs an environmental elevation map construction framework with uncertainty,and established an environmental elevation map geometric model in the robot operating system(ROS).Based on the Deeplab2 semantic segmentation network,a field t errain segmentation model was trained,and a 2.5D semantic map model framework was established.On the other hand,the tactile-based environmental characterization work was carried out.By simplifying and unifying the foot-ground interaction model under soft and hard ground,a parameterized index for characterizing the ground softness and tangential friction was proposed.Combined with geometric characteristics,a more comprehensive environmental characterization model is proposed.In terms of motion planning,we comprehensively considered the ground geometry and physical characteristics that affect the robot's passability,reconstructed the optimization goals of path planning,and achieved the optimal path planning by the graph search algorithm.Based on the idea of fault-tolerant gait,an expert planning method for free fault-tolerant gait method is proposed.To further improve the passing ability of the legged robots in the sparse foothold environment,the gait and foothold planning is processed as a sequence optimization problem.We use Monte Carlo tree search to optimize the process.Finally,two optimization methods are proposed,including fast Monte Carlo tree search and sliding Monte Carlo tree search to improve the algorithm search speed.In the experimental verification part,fisrtly the passing ability of various proposed planning methods in the sparse foothold environment was verified through random terrain experiments,and a variety of challenging artificial terrains were constructed for experimental testing.By comprehensively considering the geometric and physical characteristics,path planning and tracking experiments are carried out,and the advantages of the gait method in slippery and subsidence terrain are verified.A planning framework that considers physical characteristics combining vision and touch is constructed,and multiple sets of experimental tests are conducted.Finally,a simple whole physical experiment verification was carried out on the hexapod robot Elspider developed by Harbin Institute of Technology.Based on the work in this paper,a multi-modal environment model representation method for the legged robots,a free fault-tolerant gait planning method considering environmental fault tolerance,and a discrete contact sequence planning method based on Monte Carlo tree search are proposed.Finally,regarding the physical characteristics,a planning framework combining vision and touch is designed.This work provides a multi-modal perception and planning solution for the practical application of the legged robots.