Font Size: a A A

Research On Gait Learning Of Hexapod Robot Based On Transfer Reinforcement Learning

Posted on:2020-10-11Degree:MasterType:Thesis
Country:ChinaCandidate:K Q TangFull Text:PDF
GTID:2428330575458113Subject:Control engineering
Abstract/Summary:PDF Full Text Request
A hexapod robot is an intelligent mobile robot with multi-terrain adaptability and wide application prospects.Its movement and work need to face unstructured complex environments,which are unknown,dynamic and unpredictable.Traditional methods such as pre-programming and teleoperation need to analyze the mobility characteristics of hexapod robots in advance according to various tasks.There are many problems such as long cycle,heavy workload,low efficiency and unable to meet the needs of task diversity,which seriously restrict further applications of hexapod robots.Therefore,it is necessary to adopt robot learning to remedy the shortcomings of conventional methods such as pre-programming and improve the adaptive ability of hexapod robots to the environment.Aiming at the problem of hexapod robot mobile in unstructured environment,this paper introduces Plum-blossom Stakes landing point to model discrete landing points of robot foot and designs an all-terrain mobile algorithm for hexapod robots based on reinforcement learning.This algorithm improves the mobile ability and efficiency of hexapod robots in unstructured environments and enhances their adaptive ability to the environment.The specific contents of this paper are listed as follows.Firstly,this paper transforms human or animal behavior into mobile rules,and applies the shallow trial reinforcement learning method to the path planning of hexapod robots.This method reduces the number and time of trial and error in path selection,and improves the learning efficiency.Secondly,this paper realizes the foothold selection by the moving model of Plum-blossom Stakes,where the soft-max and ?-greedy exploration strategy are adopted for the reinforcement learning algorithm.This method improves the gait planning ability of the hexapod robot.Thirdly,in this paper,knowledge-based transfer learning and sliding window method are used to enhance the mobile strategy with better generalization performance.The robot combines the learning experience with its own motion data information,and constantly interacts with the environment to acquire new knowledge.When encountering new environments or tasks,the robot can achieve higher learning efficiency.
Keywords/Search Tags:Reinforcement learning, Hexapod robot, Transfer learning, Robot learning
PDF Full Text Request
Related items