Michael Reichel (ari)
Es spricht Armin Lederer
zum Thema: Safe Online Learning Control through Uncertainty Awareness
Abstract:
Robotic systems are increasingly designed for application scenarios, in which they need to autonomously and safely adapt their behavior to complex and dynamically changing environments. These scenarios range from rehabilitation systems, where robots need to safely interact with humans, to outdoor vehicles, which accurately need to perform tasks despite complex time-varying environmental effects. In this talk, these challenges will be addressed with a focus on the opportunities provided by integrating machine learning into high frequency, low level control loops. In particular, a framework for safe online learning control using Gaussian process regression is presented. The probabilistic foundation of Gaussian processes is exploited to obtain an explicit representation of the uncertainty of a learned model. This allows the adaptation of the robustness of control algorithms, such that safety can be effectively ensured. By establishing a direct connection between data and local model uncertainty, it is shown that simple sampling and online learning strategies can already provide strong performance guarantees for learning control systems. To realize this beneficial behavior on resource-constrained systems, computationally efficient yet guarantee-preserving approximations are proposed. The efficacy of the developed methods is illustrated using realistic simulations and real-word experiments throughout the presentation.
Dienstag, 15. April 2025, 15:00 Uhr, Curie Hörsaal
(Kaffee ab 14:30 Uhr im Raum C 325)
Alle Interessierten sind herzlich eingeladen!