MAP Prediction And Long-Term Localization For Mobile Robots Basing On ARMA Model

With the application of mobile robots in people's life and production,its long-term performance has become an important challenge in the field of robot applications,that is,robots need to adapt to the environment and work in a long-term,dynamic and open environment.Based on this,this paper mainly studies the environmental map prediction and long-term localization performance of indoor mobile robots.In the "long-term" field of indoor robots,many scholars have done relevant research.We have organized and summarized the work of the predecessors,mainly in the following two categories: 1)In terms of environmental modeling prediction,the FreMEn method proposed by Krajnik et al.believes that environmental changes are often caused by human periodic activities,and can be analyzed by the frequency of environmental changes,combined with mathematical tools such as Fourier transforms.The environment is Spatio-temporal modeled to achieve environmental prediction and adaptation.The shortcoming is that it is only suitable for environments that change periodically,and cannot be accurately constructed in the scenarios that change with other laws(such as gradual change and randomness).2)In the aspect of map updating,the dynamic grid method proposed by Tipaldi et al.considers that the change of environment is not directly related to time,and combines the fixed update trigger threshold to filter the observation data and update it on the established map.The disadvantage of this type of method is that it is based on the time independence assumption of environmental change(ie no long-term prediction information),which will result in the failure to trigger the map update due to the low observation matching,and the fixed update trigger threshold mechanism is also likely to cause the map update out of order,which affects localization robustness and accuracy.Based on the limitations of the first type of method only for periodic regular changing environments,this paper proposes the use of ARMA method for modeling long-term changing environments(can be used for environments with changing laws such as gradual change and randomness).Based on the second method,it is easy to cause the problem of map update out of order.This paper uses Bayesian filtering to combine long-term prediction with short-term observation,and proposes an updating trigger mechanism which has changeable threshold.In summary,this paper proposes an ARMA-based Map Prediction and Updating algorithm based on ARMA,namely the ArmMPU method.Finally,we perform algorithm verification through simulation and experiment.The results show that the ArmMPU algorithm has better map prediction accuracy and long-term localization performance than other methods in the complex changeble environment.