Abstract

Machine learning approaches for a few optimization problems involving lines-of-sight

Richard Tsai
University of Texas at Austin
Mathematics

I will discuss our recent work applying some deep learning strategies for solving a few optimization problems involving lines-of-sight.
The first problem is finding sparse placement of line-of-sight based sensors (LiDARs or conventional cameras) for either learning a piece of unknown domains with obstacles or putting it under surveillance.
The second problem is a zero-sum two-player surveillance-evasion game in a domain that have non-trivial obstacles obstructing the players’ line-of-sight. The two players can move freely in the obstacle free part of the domain. The game ends when the two players cannot see each other. One of the player (the pursuer) desires to keep the game going as long as possible while the other wants to end it as soon as possible.
The third problem involves visually tracking a moving adversary in an environment that have non-trivial obstacles that obstruct the pursuer’s line-of-sight, while maximizing the pursuer’s short-time surveillance of that environment.