Research Seminar on AI: Robust Execution of Abstract Task Plans on Mobile Robots

 

Dienstag, 05.10.2021, 16.00 Uhr

Despite promising advances in planning and reasoning systems, they only see little use in actual robotic environments. One reason is that solving a planning task or defining an abstract agent program is not sufficient to accomplish high-level behavior control of a robot. To execute such a plan on a robot, additional constraints of the robot platform (i.e., its hardware and low-level software components) need to be considered, which are typically ignored in the high-level behavior control.
In this talk, Till Hofmann will sketch a framework that transforms an abstract plan into an action sequence that is executable on a robot. For this purpose, a model of the robot's components is specified and constraints on that platform are defiend in metric temporal logic. Based on the platform model and constraints, an abstract plan is augmented with platform-specific actions that make sure that the platform constraints are satisfied.
In the second part, Till Hofmann will show how such a platform model is also useful during execution for error diagnosis and recovery. When executing a plan on a robot, any action may fail or have unexpected side effects, which are often not directly observable. A diagnosis is augmented as planning approach with additional information from the platform models to identify the underlying platform failure. In a production logistics scenario from the RoboCup Logistics League, Till Hofmann will present how this approach leads to a more informative diagnosis and thus a better performance of the robots.

Till Hofmann is a doctoral researcher with Prof. Lakemeyer at the Knowledge-Based Systems Group, RWTH Aachen University and he is affiliated with the DFG research training group on Uncertainty and Randomness in Algorithms, Verification and Logic (UnRAVeL).  He received his M.Sc. and B.Sc. degrees in computer science from RWTH Aachen University. His research focuses on high-level reasoning for mobile robots, in particular task planning, goal reasoning, and plan execution. As part of the team Carologistics, he has successfully participated in the RoboCup Logistics League and won the world championships in 2017 and 2019.