bei Google ScholarTitelangaben
Schwenk, Carsten ; Henrich, Dominik:
Risk-Minimizing Planning in Partially Controlled Multi-agent Systems.
In: Havlík, Štefan ; Mueller, Andreas ; Duchoň, František ; Kozák, Štefan ; Vachálek, Jan
(Hrsg.):
Advances in Service and Industrial Robotics : RAAD 2026. -
Cham
: Springer Nature Switzerland
,
2027
. - S. 449-457
. - (Mechanisms and Machine Science
; 212
)
ISBN 978-3-032-29127-1
DOI: https://doi.org/10.1007/978-3-032-29127-1_47
Abstract
During task execution in multi-agent systems (MAS), cascading effects can propagate faults across task operations and compromise safety and efficiency, necessitating strategies for mitigation. In partially controlled MAS, such as human-robot collaboration (HRC), only some agents can be actively controlled, making planning feasible for these agents only. In this paper, we present a graph-based approach to model cascading fault risk and formulate a risk-minimizing planning problem. To solve this, we introduce a dynamic Monte Carlo Tree Search (MCTS) planner that assigns turns probabilistically and accounts for uncontrolled agent strategies, enabling risk-aware operation selection for controlled agents. Experiments on synthetic tasks show consistent risk reduction over baselines across varying task sizes, speeds, and strategies.
Weitere Angaben
| Publikationsform: | Aufsatz in einem Buch |
|---|---|
| Begutachteter Beitrag: | Ja |
| Institutionen der Universität: | Fakultäten > Fakultät für Mathematik, Physik und Informatik > Institut für Informatik > Lehrstuhl Angewandte Informatik III > Lehrstuhl Angewandte Informatik III - Univ.-Prof. Dr. Dominik Henrich |
| Titel an der UBT entstanden: | Ja |
| Themengebiete aus DDC: | 000 Informatik,Informationswissenschaft, allgemeine Werke > 004 Informatik |
| Eingestellt am: | 05 Okt 2026 07:55 |
| Letzte Änderung: | 05 Okt 2026 07:55 |
| URI: | https://eref.uni-bayreuth.de/id/eprint/99586 |