Till Hofmann: Robust Execution of Abstract Task Plans on Mobile Robots
Agent formalisms such as Golog specify agent behavior in an abstract manner and ignore lower-level details. Till Hofmann’s research focuses on bridging the gap between high-level behavior specifications and their execution on a concrete robot system. To do so, Hofmann proposed to model the robot platform with timed automata and constraints based on Metric Temporal Logic. Based on this model, the abstract program is then automatically transformed into an executable action sequence. He first implemented an approach based on timed automata reachability analysis. Program and platform model are combined into a single timed automaton such that every trace in the resulting automaton corresponds to a constraintsatisfying action sequence. Experiments showed that this approach is suitable for transforming (restricted) programs and scales well to large problems. Second, in an UNRAVEL cooperation with Prof. Abrah ´ am, he is currently investigating an alternative and more expres- ´ sive approach based on MTL synthesis, which allows non-deterministic programs and an uncertain environment. Finally, the execution may still have unexpected results, e.g., due to failing actions. To deal with such failures, Hofmann uses the platform model for explanatory diagnosis. Experiments showed that platform models help to improve the diagnosis and thus resulting in better error recovery.