Solution to the Active SLAM Problem for Autonomous Vehicles Based on ADMM

Abstract

The increasing requirement of autonomy in various industry, intelligent transport systems and daily life etc has brought the great potential to the autonomous vehicles. However, to realize different levels of autonomy of autonomous vehicles, accurate and reliable environment mapping, navigation and localization techniques are essential, where the vehicles need acting efficiently such that they can localize themselves while avoiding collisions successfully. For the unknown environment exploration problem in an search and rescue scenario for example, active simultaneous localization and mapping (Active SLAM) plays a key role, whose objective can be regarded as the generation of a collision-free trajectory. A reasonable trajectory for the environment exploration may improve the efficiency of unknown environment exploring greatly. Notice that the tasks of active SLAM involve area coverage and uncertainty minimization that associated with the estimation of both robot poses and environment features. In particular, to minimize the mapping uncertainty, an uncertainty minimization index is proposed which is then converted into a non-convex constrained least-squares problem, and is then transformed into a Quadratically Constrained Quadratic Programming (QCQP)problem. Further, different from the existing semi-definite re-laxation (SDR) method, the Alternating Direction Method of Multipliers (ADMM) is utilized to solve the QCQP problem, which is suitable for large-scale structured optimization. Simulation results verify the method proposed in this paper.