Global optimal solution to SLAM problem with unknown initial estimates

The paper presents a practical approach for finding the globally optimal solution to SLAM. Traditional methods are based upon local optimization based strategies and are highly susceptible to local minima due to non-convex nature of the SLAM problem. We employed the nonlinear global optimization based approach to SLAM by exploiting the theoretical limit on the numbers of local minima. Our work is not reliant on good initial guess, whereas existing approaches in SLAM literature assume good starting point to avoid local minima problem. The paper presents experimental results on different datasets to validate the robustness of the approach, finding the global basin of attraction with unknown initial guess.