| In the 1950s,programmable robots were created to replace humans to perform repetitive tasks,but these first generation robots did not have the ability to move and were mainly used in industrial production.With the development and improvement of robotics technology,mobile robots have been invented and gradually used in many fields,such as military,rescue,service,and entertainment.However,when mobile robots are used in scenarios where humans exist,traditional navigation algorithms tend to make pedestrians feel uncomfortable because they do not consider the impact of the robot’s motion behavior on nearby pedestrians.Therefore,social navigation derived from traditional navigation is proposed to solve this problem.At present,most social navigation algorithms are proposed by improving tradi-tional navigation planning methods.Therefore,they may still have some typical prob-lems,such as ignoring the nonholonomic and dynamics constraints of the differentially driven mobile robot,easily falling into local minima and high computational cost.To solve these problems,this dissertation designs the algorithm based on the basic frame-work of Dynamic Window Approach,and proposes indoor social navigation algorithm for a differentially driven mobile robot based on multi-objective evaluation.The al-gorithm not only considers the constraints of the differentially driven robot and local minimum problem,but also has the advantages of naturalness,comfort and real-time.Specifically,the main contents of this dissertation are comprised as follows.(1)Research two popular gap extraction algorithms,CG and AG.For the gap mis-judgment defects of these algorithms,this dissertation proposes NG algorithm to im-prove it.The three algorithms are compared through simulation experiments and the results show that the accuracy of NG algorithm is significantly higher than the other two algorithms.(2)Research five typical line extraction algorithms.Additionally,real-time Split-and-Merge algorithm,EKF-based algorithm and PFE algorithm are used for line ex-traction test on the lidar point cloud data with Gaussian noise.The results show that the performance of these three algorithms is similar in terms of accuracy,but Split-and-Merge algorithm has the fastest extraction speed and is least affected by noise.(3)Research the basic form of the social force model and its improved forms,and test the influence of different social force model parameters on pedestrians or robots in terms of trajectory length and running time in a simulation environment.In addition,social work is introduced and used to quantify the comfort of navigation,and the effec-tiveness of this metrics is verified by comparing the performance of different navigation algorithms.(4)The social navigation algorithm is proposed based on a combination of global planner and local planner.Specifically,global planner is employed to provide a global reference path.The local planner is designed based on the above research and the idea of multi-objective evaluation.By comparing the performance of the robot’s predicted trajectories in terms of target heading,obstacle avoidance and pedestrian comfort,the linear and angular velocities corresponding to the optimal trajectory are selected as the control input of the robot.The simulation results show that multi-objective evaluation based social navigation algorithm proposed in this dissertation can further improve the smoothness of the trajectory on the basis of Dynamic Window Approach,and can ef-fectively lessen the possibility of ending in local minima.Moreover,the real-time per-formance and comfort of navigation have been significantly improved.The feasibility of the algorithm is also verified by physical experiments. |
