Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | Agents and Artificial Intelligence |
| Untertitel | 13th International Conference, ICAART 2021, Virtual Event, February 4–6, 2021, Revised Selected Papers |
| Herausgeber/-innen | Ana Paula Rocha, Luc Steels, Jaap van den Herik |
| Herausgeber (Verlag) | Springer Science and Business Media Deutschland GmbH |
| Seiten | 158-180 |
| Seitenumfang | 23 |
| ISBN (elektronisch) | 978-3-031-10161-8 |
| ISBN (Print) | 9783031101601 |
| Publikationsstatus | Veröffentlicht - 19 Juli 2022 |
| Veranstaltung | 13th International Conference on Agents and Artificial Intelligence, ICAART 2021 - Virtual, Online, Österreich Dauer: 4 Feb. 2021 → 6 Feb. 2021 |
Publikationsreihe
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Band | 13251 LNAI |
| ISSN (Print) | 0302-9743 |
| ISSN (elektronisch) | 1611-3349 |
Abstract
Using multi-agent reinforcement learning to find solutions to complex decision-making problems in shared environments has become standard practice in many scenarios. However, this is not the case in safety-critical scenarios, where the reinforcement learning process, which uses stochastic mechanisms, could lead to highly unsafe outcomes. We proposed a novel, safe multi-agent reinforcement learning approach named Assured Multi-Agent Reinforcement Learning (AMARL) to address this issue. Distinct from other safe multi-agent reinforcement learning approaches, AMARL utilises quantitative verification, a model checking technique that guarantees agent compliance of safety, performance, and non-functional requirements, both during and after the learning process. We have previously evaluated AMARL in patrolling domains with various multi-agent reinforcement learning algorithms for both homogeneous and heterogeneous systems. In this work we extend AMARL through the use of deep multi-agent reinforcement learning. This approach is particularly appropriate for systems in which the rewards are sparse and hence extends the applicability of AMARL. We evaluate our approach within a new search and collection domain which demonstrates promising results in safety standards and performance compared to algorithms not using AMARL.
ASJC Scopus Sachgebiete
- Mathematik (insg.)
- Theoretische Informatik
- Informatik (insg.)
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
Agents and Artificial Intelligence : 13th International Conference, ICAART 2021, Virtual Event, February 4–6, 2021, Revised Selected Papers. Hrsg. / Ana Paula Rocha; Luc Steels; Jaap van den Herik. Springer Science and Business Media Deutschland GmbH, 2022. S. 158-180 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 13251 LNAI).
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Assured Deep Multi-Agent Reinforcement Learning for Safe Robotic Systems
AU - Riley, Joshua
AU - Calinescu, Radu
AU - Paterson, Colin
AU - Kudenko, Daniel
AU - Banks, Alec
N1 - Funding Information: Supported by the Defence Science and Technology Laboratory.
PY - 2022/7/19
Y1 - 2022/7/19
N2 - Using multi-agent reinforcement learning to find solutions to complex decision-making problems in shared environments has become standard practice in many scenarios. However, this is not the case in safety-critical scenarios, where the reinforcement learning process, which uses stochastic mechanisms, could lead to highly unsafe outcomes. We proposed a novel, safe multi-agent reinforcement learning approach named Assured Multi-Agent Reinforcement Learning (AMARL) to address this issue. Distinct from other safe multi-agent reinforcement learning approaches, AMARL utilises quantitative verification, a model checking technique that guarantees agent compliance of safety, performance, and non-functional requirements, both during and after the learning process. We have previously evaluated AMARL in patrolling domains with various multi-agent reinforcement learning algorithms for both homogeneous and heterogeneous systems. In this work we extend AMARL through the use of deep multi-agent reinforcement learning. This approach is particularly appropriate for systems in which the rewards are sparse and hence extends the applicability of AMARL. We evaluate our approach within a new search and collection domain which demonstrates promising results in safety standards and performance compared to algorithms not using AMARL.
AB - Using multi-agent reinforcement learning to find solutions to complex decision-making problems in shared environments has become standard practice in many scenarios. However, this is not the case in safety-critical scenarios, where the reinforcement learning process, which uses stochastic mechanisms, could lead to highly unsafe outcomes. We proposed a novel, safe multi-agent reinforcement learning approach named Assured Multi-Agent Reinforcement Learning (AMARL) to address this issue. Distinct from other safe multi-agent reinforcement learning approaches, AMARL utilises quantitative verification, a model checking technique that guarantees agent compliance of safety, performance, and non-functional requirements, both during and after the learning process. We have previously evaluated AMARL in patrolling domains with various multi-agent reinforcement learning algorithms for both homogeneous and heterogeneous systems. In this work we extend AMARL through the use of deep multi-agent reinforcement learning. This approach is particularly appropriate for systems in which the rewards are sparse and hence extends the applicability of AMARL. We evaluate our approach within a new search and collection domain which demonstrates promising results in safety standards and performance compared to algorithms not using AMARL.
KW - Assurance
KW - Assured Multi-Agent Reinforcement Learning
KW - Deep Reinforcement Learning
KW - Multi-Agent Reinforcement Learning
KW - Multi-Agent Systems
KW - Quantitative verification
KW - Reinforcement Learning
KW - Safe Multi-Agent Reinforcement Learning
KW - Safety-critical scenarios
UR - http://www.scopus.com/inward/record.url?scp=85135062258&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-10161-8_8
DO - 10.1007/978-3-031-10161-8_8
M3 - Conference contribution
AN - SCOPUS:85135062258
SN - 9783031101601
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 158
EP - 180
BT - Agents and Artificial Intelligence
A2 - Rocha, Ana Paula
A2 - Steels, Luc
A2 - van den Herik, Jaap
PB - Springer Science and Business Media Deutschland GmbH
T2 - 13th International Conference on Agents and Artificial Intelligence, ICAART 2021
Y2 - 4 February 2021 through 6 February 2021
ER -