Font Size: a A A

Trajectory Planning And Control Of Mobile Grasping Robot Based On Reinforcement Learning

Posted on:2022-10-09Degree:MasterType:Thesis
Country:ChinaCandidate:W Y ChenFull Text:PDF
GTID:2518306536967359Subject:Engineering
Abstract/Summary:PDF Full Text Request
With the rise of artificial intelligence technology,the birth of intelligent robots has greatly facilitated all aspects of human beings.The intelligent mobile grasping robot that integrates intelligent voice,SLAM technology,target detection and tracking algorithms,and reinforcement learning multi-modal interaction must have broad development prospects and practical significance.This paper takes the intelligent mobile grasping robot as the research object to study the problem of target tracking and trajectory planning in an unknown environment.The main contents are as follows:(1)Aiming at the problem of target tracking in complex background environments,a new adaptive layered resampling mutation particle filter algorithm is designed to solve the problem of target loss caused by interference in complex recognition situations.In resampling,a hybrid proposal distribution method is proposed to sample the next generation of particles,which solves the problem of severe degradation of particles;optimization algorithm to set a high weight threshold and a low weight threshold,and compare the weight of the particles into Three groups:The weight optimization operation is used to optimize the corresponding weights,which solves the problems of accuracy and long running time;for the problem of diversity loss caused by particle aggregation after adaptive resampling of the algorithm,we adopt a mutation method for particles enhance the diversity of particles.(2)Aiming at the grasping problem of the robotic arm in an unknown environment,combined with SLAM and voice interaction technology,a robotic arm trajectory planning algorithm based on deep reinforcement learning to solve the problem of optimal posture trajectory planning of the robotic arm in an unknown environment.First,an experience separation mechanism is designed to distinguish between positive and negative experiences and store them in different experience buffers;second,a variable ratio replay mechanism is designed to describe the positive and negative experiences generated to the optimization level of the decision-making model,variable ratio mechanism can adjust the ratio of positive experience and negative experience in real time.Finally,a priority sampling mechanism for the experience playback area is proposed,which uses a time-series differential deviation method to measure each learning value of this experience.The experience in the experience pool by the absolute value of the timing difference deviation,replay those experiences with high deviation more frequently,avoid making wrong behaviors again,improve the overall performance,accelerate the convergence speed of the algorithm.Finally,the feasibility of this algorithm is verified by simulation and experiment.(3)Finally the construction of the entire intelligent mobile grasping robot experimental platform,introduced the hardware composition and software structure of the entire platform.designed a voice-driven control system,analyzed the voice recognition process,and the mobile robot voice recognition system The acoustic model used in the article is explained,the deep neural network-hidden Markov model is described,and experimental tests are carried out.the grasping control system is designed,and the specific method including coordinate transformation is introduced.Finally,the process and results of the whole experiment are shown in detail.
Keywords/Search Tags:Trajectory planning, reinforcement learning, particle filtering, speech recognition, multimodal interaction
PDF Full Text Request
Related items