Forschungsthemen
[MA] Grammar-based Behavior Trees for Industrial Robotics
Recently, behavior trees (BTs) emerged as a popular abstraction to describe tasks in robotics, due to their intuitive semantics and concise specification. BTs can be constructed manually or obtained by machine learning or rule-based systems. This work focuses on a special case of rule-based systems, grammars, as a specification language for BTs. Grammars have been used to describe robot behavior, e.g., in the form of motion grammars (MGs). However, this ap- proach is unable to express concurrent behavior and lacks an equivalent of the fallback mech- anisms present in BTs. On the other hand, MGs can not only express procedurally constructed trees, but also have a solid concept for data flow (as opposed to BTs), but have drawbacks in specification complexity. Therefore, this work investigates the extension of motion grammars to incorporate the features a behavior trees, resulting in a merged concept.
Betreuer: Johannes Mey