Research On Optimization Method Of Modular Robots’ Configuration And Motion Strategy

In recent years,with the rapid development of artificial intelligence and robotics,the application scenarios of robots have become more and more complex,gradually transitioning from structured environments to unstructured environments.Traditional robots with a single function and a fixed structure have limited task detection capabilities in complex unstructured environments.In contrast,modular robots have strong capabilities due to their reconfigurability,self-repair,and adaptability.Environment and task adaptability,research value and application potential are huge.In addition,limited by factors such as experimental cost and experimental period,the design,testing and evaluation of modular robots are mainly carried out in the simulation environment.The existing open-source simulation experimental platforms cannot meet the requirements of fast and efficient verification of reconstruction planning methods in complex scenarios.need.Therefore,it is urgent to carry out theoretical research on the optimization method of modular robot configurationmotion strategy and development of simulation experimental platform to improve the autonomous decision-making ability and adaptive ability of modular robot configuration-motion strategy.Focusing on the detection requirements of modular robots in multi-task scenarios,this paper starts from two aspects: theoretical research and simulation environment development,and systematically studies its configuration-motion strategy optimization problem and simulation experiment platform development.The main research contents are as follows:(1)Motion strategy optimization method based on deep reinforcement learning.Considering the problems of high sampling cost and low sample utilization in the existing research on the motion optimization of modular robots,how to effectively use the failure experience in the exploration process has become one of the difficult problems in this field.For this reason,based on the classic TD3 algorithm in deep reinforcement learning,this paper proposes a motion optimization algorithm based on the TD3 algorithm(H-TD3)combined with the post-experience replay mechanism.Increase the number of successful experiences in the experience pool,accelerate the learning process of the robot,and improve the efficiency of the motion optimization of the modular robot.(2)A configuration-motion co-optimization method based on evolutionary computation.Starting from a fixed configuration to optimize the motion strategy of the robot,the search space is reduced to a certain extent,but the final optimization effect may be limited by the choice of the initial configuration.Therefore,the configuration-motion synergistic optimization of the modular robot is carried out,and how to effectively use environmental information in the optimization process has become one of the key research contents in this field.This paper firstly introduces a co-optimization method based on evolutionary computing method,and designs a general dynamic coding method for modular robots and a variety of evolutionary operators.On this basis,in order to reduce the computational complexity of joint optimization,a co-evolution method based on terrain prior is proposed,combined with the terrain distribution information obtained by the semantic segmentation method to construct a better initial robot population,and continuously optimize and update to obtain a group that can efficiently complete detection.The robot entity for the task.The proposed method realizes the joint optimization of environment,configuration and motion,and improves the solution efficiency of collaborative optimization problems.(3)Development of modular robot reconstruction simulation experiment platform.Aiming at the problems of poor scalability of the existing platform and difficulty in effectively simulating complex environments,the We Volve simulation experiment platform is built based on the Webots simulator,which provides a solution framework for motion optimization and configuration-motion co-optimization,and supports different types of complex scenarios.Reconstruction simulation experiment of modular robot.It provides a new,efficient and multifunctional research tool and platform for the field of modular robotics research,and has important engineering value for the further development of this field.