Fast Localization Method Of An Indoor Autonomous Mobile Robot Based On Particle Filter

Localizaiton refers to the robot's position in the map.It is one of the important techniques for the mobile robot to achieve autonomous motion,and it is the premise for the robot to perform the corresponding tasks.Particle filtering is an effective global localizaiton method.By spreading particles in the global map as multiple guesses of the robot pose,the sensor data is compared with the particle's environment in the map to obtain the confidence of the particles.However,in order to make the particle filter show higher localizaiton accuracy,it is necessary to maintain a larger number of particles,which will cause the calculation speed to decrease.If you want to maintain the localizaiton speed,the computational resource burden of the robot will increase,so how to improve the calculation speed of the particle It is important to reduce the cost of particle filter localizaiton.In this paper,the particle filter localizaiton algorithm is slow in maintaining a large number of particles.It is proposed to use a high-precision odometry to improve the accuracy of particle distribution,and to improve the weighting process and improve the speed of the weighting process.The main work of this paper is as follows:(1)In order to make the particle distribution more accurate,this paper uses the encoder integrated with the IMU to replace the single encoder odometry,and then adds the travel noise generated by the robot motion model to move the particles,which improves the accuracy of particle distribution.Ensure that the particles cover the true pose of the robot,so the particle distribution range is reduced to a certain extent,and the number of particles used can be reduced.(2)In the subsequent stage of calculating the particle weight,the ray casting algorithm is needed to calculate the nearest obstacle distance around the particle.The use of this algorithm on each particle creates the disadvantage of taking too long a computation.Therefore,this paper proposes a lookup table method,which executes the ray casting algorithm in advance and stores the result in the lookup table.On this basis,an improved lookup table method is proposed.This lookup table method can maintain basically the same acceleration performance compared with the conventional search,and can reduce the pre-calculation time and reduce the memory occupied by the lookup table.Finally,for the particle filter with parallelization structure,it is also proposed to use the GPU(graphics processor)to parallelize the weighted parts of the particle,and combined with the improved lookup table method to further improve the operation speed.(3)Experiments based on ROS(robot operating system)show that on the basis of using high-precision odometry to distribute particles,when using the look-up table method for particle weighting,the weighting calculation speed can be increased by several tens of times,but the cost of this method Is the lookup table memory is too large.The improved lookup table method solves this drawback.When using GPU operations,the particle update speed is also faster,but the effect is not as good as the lookup table method.When the GPU combines the improved lookup table with the fastest speed,the weight update of 5000 particles takes only about 10 ms.The method of this paper makes the particle filtering speed greatly improved,and this method has lower requirements on computing resources,which can reduce the robot cost and improve the robot positioning speed or accuracy.