Andrew Stout

Behavior trees are a mechanism for representing and abstracting complex behavior for autonomous agents. While they originated in the video game industry, they have been steadily gaining popularity in robotics over the past several years. In contrast to other widely used approaches such as finite-state machines, behavior trees offer several benefits in the structure they enforce, including a relatively concise standard modeling language, ease of modification and extension, and verification guarantees. This presentation will begin with an introduction to behavior trees, which covers the formalism, basic mechanics, and common idioms for robotics applications. We will then contrast behavior trees with alternative approaches and detail the main software packages and resources available for getting started. Finally, we will present some lessons learned from experience applying behavior trees in various robotics projects, spanning a range of application areas and academic versus commercial uses.