Research On Mobile Robot Positioning And Path Planning Method Based On Multi-Sensor Fusion

Ground mobile robots are widely used in industry and daily life due to their exceptional platform stability and flexible movement.At present,when mobile robots are facing more complex working conditions and tasks,the requirements for high-precision positioning and effective movement planning are also increasingly improved.In terms of positioning,the system with a single sensor can no longer guarantee the accuracy and sustainability,and it is easy to cause the loss and deviation of the information when disturbed.In terms of path planning,global planning methods cannot respond to unknown environments or their dynamic changes,and also cannot guarantee the real-time planning.Due to the lack of guidance from global information,local planning methods focus more on instant obstacle avoidance,and are short of the orientation of achieving the target,making it prone to problems such as being far from the target heading or getting stuck in reciprocating hovering.Aiming at the above problems,this thesis has carried out the research of a combined positioning method based on multi-sensor fusion and a path planning method combining global and local approaches,using a mobile robot as a platform.The research content and results of this thesis are as follows:(1)In the research of combined positioning methods,aiming at the dependence of GPS on satellite signals and the time accumulated error of inertial navigation,the inertial navigation method combining the two was studied.Based on the analysis of the update algorithm and error model of inertial navigation,a GPS/INS combined positioning system based on position/speed was implemented.Aiming at the limitation that Kalman filter cannot be applied to nonlinear systems,a combined filter based on unscented Kalman filter was implemented.The combined positioning system was tested in a simulation environment.The results showed that the positioning error of GPS/INS combined positioning system can be controlled within 2m,the speed error can be controlled within 0.3m/s,the positioning accuracy is about 0.5m,and the speed accuracy is about 0.05m/s.Significant improvement is achieved in positioning accuracy compared to the positioning system with a single sensor.(2)In the problem of local obstacle avoidance for mobile robots,research on the vector field histogram method was carried out.Aiming at the loopholes in the motion model and direction strategy of the original algorithm and its sensitivity to thresholds,new motion model and direction strategy were supplemented,and an adaptive threshold adjustment strategy was proposed to enable the algorithm to dynamically adjust threshold parameters based on environmental characteristics.The obstacle avoidance and planning functions of the improved algorithm were tested in simulated static and dynamic obstacle environments.The results showed that compared to the original algorithm,the improved algorithm makes the path of obstacle avoidance smoother,has a higher success rate of reaching the target,and can shorten the length of the path by10% to 15%.Also,the improved algorithm has better responses to dynamic obstacles.(3)In global path planning,deep learning network algorithm combining neural networks and reinforcement learning was studied for high-dimensional problems when reinforcement learning is applied to complex continuous problems.In order to accelerate the convergence speed of training and improve the robustness of state updates,the advantageous function and dueling network were introduced to optimize the network structure,enabling the algorithm to judge the superiority when selecting actions.The optimized algorithm was tested in simulation environments.The results showed that compared to the original algorithm,the optimized algorithm can adopt a superior action strategy when planning paths,and can obtain higher average rewards.Also,the optimized algorithm has good adaptability to larger maps,and can maintain the planning accuracy rate of over 92%.The combined positioning method and hybrid path planning method were systematically integrated,and the feasibility of the system was verified through physical simulation platforms and visualization tools.In simulation experiments,the hybrid path planning method can more clearly judge the distribution of obstacles,and make the total reward of planning results reach convergence states.Finally,the positioning and path planning functions of mobile robots were implemented in real environments.