A Review of Learning-Based SLAM Approaches of Autonomous Unmanned Vehicles (AUV)

Artificial Intelligence (AI) is becoming a hot topic in the field of robotic research in the last few decades. Autonomous Unmanned Vehicles (AUV) are being used for different tasks like rescue, search, monitoring, aerial operations as well as underwater operations even AUV can aid where human reachability is impossible. Localization, tracking, and mapping are fundamentals of an autonomous system. The main problem of AUV, which attracts researchers, is the simultaneous localization and mapping, where no external positioning source is available like GPS. There are many proposed techniques and algorithms which can be used to solve this problem of AUV like GPS (Global Positioning System), Motion Capture System (MCS), Visual System, etc. with limitations. Some probabilistic solutions like Graph SLAM, EKF based SALM, and Fast SLAM are also available for this problem. EKF based SLAM is used for the non-linear model but has different issues (like inconsistency) when the map becomes large and complex. It does not work well with the non-Gaussian distribution. Fast SLAM algorithm is used with the non-Gaussian distribution. It can provide high speed computation and good accuracy but has issues during the resampling processes like particle depletion and degeneracy. On the other hand, Graph based SLAM can deal with large and complex maps and can process a large number of landmarks accurately and it can perform much better than EKF and Fast SLAM.