Towards Reactive and Robust Manipulation Tasks using Behavior Trees

The applications of manipulation tasks will soon move from caged robots in assembly lines to the more challenging household and collaborative ones that have to operate in dynamic and unpredictable environments. Due to the uncertainty of such environments, it is preferable to perform tasks in a reactive and robust fashion. Classic task execution architectures fail to do so in reasonable complexity, which suggests considering alternative execution architectures. In this paper, we outline how to achieve reactive and robust manipulation by encoding the task as a Behavior Tree (BT). To demonstrate our approach, we illustrate an example of a manipulation task and how the BT deals with external disturbance and failures at the task level. We provide a preliminary comparison with Finite State Machines and State Charts.