EFFICIENT GLOBAL LOCALIZATION BY SEARCHING A BIT-ENCODED GRAPH

Abstract Global localization is the problem of finding a totally unknown robot pose given a known map and a set of observations relative to the robot. This is the combinatorial correspondence problem between map landmarks and observations, and its complete solution is an NP-hard problem that makes it computationally intractable when the size of the map increases. We propose the use of bit encoded graphs that take advantage of bit parallelism when solving instances of the deeply studied in graph theory Maximum Clique Problem (MCP). We have implemented and compared our solution with standard branch and bound heuristic search over the space of possible associations. Although the computational complexity of our algorithm is still exponential because it is complete, our results prove an impressive reduction of two orders of magnitude in required processing time thanks to the efficient underlying model, making it possible to deal with previously computationally intractable scenarios. Moreover, it is the first time (up to our knowledge) that this kind of technique is used in the robotics domain and it could possibly be applied to other problems such as path planning.