Research On Decision Making And Control Of Mobile Robots In Complex And Dynamic Environments

The traditional mobile robot autonomous navigation system divides the entire complex autonomous navigation problem into multiple sub-problems such as localization,perception,planning,decision-making,and control.In recent years,the research of mobile robots has gradually moved from demonstration in a simple experimental environment to practical application in real scenarios.The working environment it faces also ranges from two-dimensional structured simple static environments to three-dimensional unstructured complex dynamic environments.However,the traditional modular-based autonomous navigation system,especially its decision-making and control module,the degree of automation and intelligence of the system is obviously insufficient in complex autonomous navigation tasks.In order to realize the migration of mobile robots from simple experimental environments to actual complex tasks,and comprehensively improve the intelligence of mobile robots,especially for the flexibility of the decision module and the real-time performance of the control module.This study starts from three typical tasks scenarios,and analyzes the related decision-making and control problems involved in robots.The specific research contents are described as follows:(1)For the task of automated visual surveillance,this study uses multiple robots to achieve real-time collaborative and precise perception in a large-scale surveillance scene.Then the game model for the complex interaction in the surveillance scene is modeled based on the game theory,including cooperation and confrontation among multiple agents.Finally,by cooperating with the motion control algorithm,the realtime dynamic hunting of suspicious intrusion targets is realized.By constructing a set of deeply integrated perception,decision-making and control algorithms,the algorithm overcomes the shortcomings of the traditional fixed camera-based visual surveillance system in mobility and intelligence,and tt also demonstrates the potential of game-based decision-making strategies to flexibly deal with complex task scenarios in a modular scheme.(2)For the task of multi-robot formation control,this study solves the traditional formation control problem from the perspective of numerical optimization.By introducing nonlinear optimization algorithm and combining with traditional control methods,a variety of multi-robot cooperative formation algorithms with practical application background are designed.Finally,with the help of the fast convergence characteristics of the optimization algorithm and the automatic parameter search,the collaborative formation algorithm proposed in this study can be automatically tuned in real time according to specific scenarios,which improves the real-time performance and environmental adaptability of mobile robots in actual formation tasks.(3)For the task of automation navigation,this study from a data-driven perspective investigated the problem of decision control strategy learning of mobile robots.By constructing a variety of strategy learning datasets with both spatiotemporal information,and designing a novel end-to-end decision control strategy deep learning model based on spatiotemporal information.Finally,the deep integration of perception,planning,decision-making,and control modules enhances the learning ability and scene transfer ability of the automatic navigation system in complex environments.Finally,Based on the above improvements and innovations,this study improves the flexibility of decision-making and the real-time performance of control in existing modular-based navigation schemes in surveillance,cooperative formation tasks from the perspectives of game decision-making and optimal control,and from a spatiotemporal data-driven perspective improves the perception,learning,and reasoning capabilities of the learning-based navigation scheme in autonomous driving tasks.It provides a reference for mobile robots to complete other practical tasks in complex dynamic scenarios.