Midterm colloquium Liam Bosland

24 September 2020 13:00 till 13:30 - Location: no audience - By: DCSC

"Accelerated ADMM for Distributed Pose Graph Optimization in SLAM problems"

In order to create fully autonomous robots, they should be able to localize themselves in unknown environments. One way to approach this problem is through pose graph optimization (PGO). In PGO a robot creates a graph in which its poses are represented by nodes and the measurements are represented by edges between these nodes.

Minimizing the mismatch between the measurements and the poses over this graph results in an accurate representation of the environment and the trajectory of the robot. A distributed approach to PGO is necessary as graphs grow larger and more robots are added to the system. Fast optimization methods with low memory consumption are necessary to allow for these types of systems. The thesis work aims to contribute an accelerated implementation of the alternating direction method of multipliers (ADMM) for distributed PGO. Implementing this algorithm for distributed PGO allows for a possible increase in performance when compared to the related works. The obtained results in implementation and future directions of the thesis work will be discussed during this midterm colloquium.

Supervisor:
Dr.ir. T. Keviczky