A survey of algorithmic methods for partially observed Markov decision processes
Annals of Operations Research
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
Proceedings of the Twelfth International Florida Artificial Intelligence Research Society Conference
Hybrid multi-agent architecture as a real-time problem-solving model
Expert Systems with Applications: An International Journal
Comparing real-time and incremental heuristic search for real-time situated agents
Autonomous Agents and Multi-Agent Systems
Online planning algorithms for POMDPs
Journal of Artificial Intelligence Research
The ARTS real-time agent architecture
LADS'09 Proceedings of the Second international conference on Languages, Methodologies, and Development Tools for Multi-Agent Systems
An overview of cooperative and competitive multiagent learning
LAMAS'05 Proceedings of the First international conference on Learning and Adaption in Multi-Agent Systems
| Hi-index | 0.00 |
Distributed systems based on cooperative multi-agents have been used in a wide range of application domains. However, the need for real-time processing in large and dynamic search spaces has led to new challenges. In addition to the constraints in time and computational resources, the agents have to operate under highly dynamic conditions in complex environments. Finding optimal solutions within time constraints may not be always possible. Anytime algorithms have shown great promise in providing approximate solutions. The quality of these intermediate/partial solutions depends on the amount of computational resources available for processing. The key insight that we describe in this paper is that anytime algorithms can be leveraged in a partial processing paradigm where the partial solutions can be used to quickly identify potential solutions and thereby efficiently utilize resources, even under dynamic conditions. The partial solutions can also be used for a coarse grained categorization of large search spaces that can support a mix of explorative and exploitative agent behaviors. We will describe how explicit modeling of the dynamism using a simple but unique search space model can help agents adapt to the changing information space. We describe a generic multi-agent framework that leverages our search space model while modeling various aspects of agent behavior such as candidate selection, agent interactions, etc. This framework can be used to design agents to work with partial processing in various application domains. We will develop suitable testbeds, simulation experiments, algorithms and performance metrics to validate the framework.