Bio-Inspired Bathymetric SLAM System Considering Invalid Loop Closures

Bio-inspired systems, such as RatSLAM, which are modeled on hippocampal functions, have proven to be accurate and reliable solutions for the SLAM problem. However, in bathymetric SLAM, the nearly flat seabed terrain makes the occurrence of false loop closures unavoidable. These false loop closures can cause perceptual aliasing, leading to multiple position estimates for the vehicle. Conventional bio-inspired systems lack the higher-order capacities of the medial prefrontal cortex (MPFC) required to maintain multiple position hypotheses concurrently. To address this perceptual aliasing problem induced by false loop closures, we propose a bio-inspired bathymetric SLAM system endowed with reasoning capabilities. Unlike traditional bio-inspired systems, the proposed system can maintain a set of activity packets simultaneously to represent perceptual ambiguity. Similarities between the current submap and submaps stored in historical local view cells not only activate pose cells but also generate inhibitory signals to suppress competing activity packets. Additionally, we introduce an attenuation mechanism for these inhibitory signals when mismatches occur between the current submap and historical submaps. The performance of the proposed bio-inspired system is evaluated through simulation experiments using field data.