Loss Function Regularization on the Iterated Racing Procedure for Automatic Tuning of RatSLAM Parameters

AbstractSimultaneous localization and mapping (SLAM) is a fundamental problem in mobile robotics. Among the solutions to solve this problem, the RatSLAM is a SLAM algorithm inspired by the spatial navigation system of rodents. The RatSLAM has a set of parameters that must be adjusted for each new environment to generate a reasonable map. A proposed solution to tune these parameters has been presented in the form of a manual trial-and-error algorithm. This algorithm guides the tuning only in part of the environment, despite the adjustment being suitable for the entire environment. In addition, recent work has proposed an automatic parameter tuning solution using the Iterated Race (irace) and Iterative Closest Points (ICP). However, this automatic solution provides parameters values that might be suitable only for the environment places in which they were adjusted, i.e.the adjustment might not be adequate for the entire environment. This work proposes a regularisation of the automatic algorithm objective function to incorporate the advantages of the manual solution into it. The proposed process uses only part of a virtual environment to find the parameters for the RatSLAM. Then, these parameters are tested in new places and the entire environment. The results have shown that the parameters found by the approach can generalize for new areas, as well as be suitable to map the entire environment.KeywordsSLAMRatSLAMParameter tuning