A Game of Bundle Adjustment -- Learning Efficient Convergence

TL;DR

This paper introduces a reinforcement learning approach to optimize the damping factor in bundle adjustment, significantly reducing iterations and improving efficiency in localization and mapping tasks.

Contribution

It presents a novel formulation of bundle adjustment as a reinforcement learning problem to learn optimal damping factors, enhancing convergence speed.

Findings

Reduces the number of iterations for convergence

Applicable to both synthetic and real-world data

Can be integrated with existing acceleration methods

Abstract

Bundle adjustment is the common way to solve localization and mapping. It is an iterative process in which a system of non-linear equations is solved using two optimization methods, weighted by a damping factor. In the classic approach, the latter is chosen heuristically by the Levenberg-Marquardt algorithm on each iteration. This might take many iterations, making the process computationally expensive, which might be harmful to real-time applications. We propose to replace this heuristic by viewing the problem in a holistic manner, as a game, and formulating it as a reinforcement-learning task. We set an environment which solves the non-linear equations and train an agent to choose the damping factor in a learned manner. We demonstrate that our approach considerably reduces the number of iterations required to reach the bundle adjustment's convergence, on both synthetic and real-life scenarios. We show that this reduction benefits the classic approach and can be integrated with other bundle adjustment acceleration methods.