October 2017: Guaranteed computation of robot trajectories


By Simon Rohou, Luc Jaulin, Lyudmila Mihaylova, Fabrice Le Bars, Sandor Veres
Guaranteed Computation of Robot Trajectories
Robotics and Autonomous Systems
Volume 93, July 2017, pp76-84
DOI: 10.1016/j.robot.2017.03.020

Summary: Intelligent systems operate in unknown environments and often use sensors whose characteristics are not accurately known. In such cases a safe option is to only impose upper bounds on the sensor measurement errors. In order to solve the challenging task of localisation of robots under bounded sensor errors, this work proposes an efficient guaranteed estimation method for robot states and applies it to mobile robotics.

Although the problem of robot localisation has been widely studied, not many approaches can deal with large measurement errors or unknown measurement characteristics. The developed guaranteed solution is a novel set-membership method for non-linear dynamic systems. A constraint programming approach is applied by enclosing state trajectories of the robot into specific domains called tubes. We provide an operator, a so-called differential contractor, to reliably reduce these domains according to the defined constraints.

This framework is simple to use and more general than existing approaches dealing with guaranteed integration. Besides, the method can easily handle real data. We demonstrate its efficiency through a realistic underwater experiment involving an autonomous vehicle. The source code of the framework is freely available online.