OntheObservability andObservability Analysis ofSLAM
Wijerupage Sardha Wijesoma · 2006
Simultaneous localization andmappingproblem formobilerobots hasreceived considerable attention overthe last decade. Thewidely usedformulation oftheSLAM problem hasbeentheaugmented state approach inanestimation theoretic framework. Although, manyrelated issues ofSLAM suchas computational complexity, loopclosing anddataassociation have received much attention, theobservability issuehaslargely remained ignored. Systemobservability isanimportant aspect in anystate estimation problem. Observability analysis provides for understanding of thefundamental limits of thesolution obtainable, regardless ofprocess andmeasurement noises. The standard world-centric SLAM formulation isahighly non-linear system. Thusthedirect useoflinear observability tools and criteria intheanalysis ofitsobservability yields incorrect and inconsistent results. Inthispaperan appropriate methodof analysis oftheobservability ofnon-linear systems isapplied to investigate theproperties ofthestandard SLAM formulation. Contrary topopularbelief, itisshownthroughtheoretical analysis thatthestandard 2D planarworld-centric SLAM formulation involving odometry inputs forrobotspeedand heading, andrange/bearing measurements tofeatures inthe environment isunobservable. Itisalsoshownthatforthesystem tobe observable, itrequires atleast twoabsolutely known feature pointpositions, thusquestioning theverymeaning implied bySLAM.Theanalytical results thusestablished are verified through simulations. IndexTerms-Localization, SLAM&Mobile Robot.