Towards Constant-time Approximation of Global Uncertainties in Local Bundle Adjustment

OSSENKOPF, MARIE; NITSCHE, MATIAS; PESSACG, FACUNDO; GEIHS, KURT; CASTRO, GASTON; FISCHER, THOMAS

Buenos Aires

Workshop; Outreach Workhsop of International Symposium on Experimental Robotics; 2018

Proper recovery of the uncertainty estimates of the robot's trajectory and mapped landmarks allows performing uncertainty-aware tasks such as active planning and navigation. Although globally consistent uncertainties are readily available when computing a batch-optimized solution of the SLAM problem, this solution becomes prohibitively expensive as the mapped area increases. This is typically overcome by optimizing a locally bounded portion of the map at each step, but the availability of consistent uncertainties is lost in the process. The current work proposes a way to perform local optimization over an arbitrarily selected area of the mapped environment while still recovering a consistent approximation of the global uncertainties with complexity proportional to the size of the local map.