Multi-agent reinforcement learning: independent vs. cooperative agents
Readings in agents
Multiagent Systems: A Survey from a Machine Learning Perspective
Autonomous Robots
Evolving neural networks through augmenting topologies
Evolutionary Computation
Evolving Beharioral Strategies in Predators and Prey
IJCAI '95 Proceedings of the Workshop on Adaption and Learning in Multi-Agent Systems
Reinforcement learning agents with primary knowledge designed by analytic hierarchy process
Proceedings of the 2005 ACM symposium on Applied computing
Hi-index | 0.04 |
The prey-predator pursuit problem is a generic multi-agent testbed referenced many times in literature. Algorithms and conclusions obtained in this domain can be extended and applied to many particular problems. In first place, greedy algorithms seem to do the job. But when concurrence problems arise, agent communication and coordination is needed to get a reasonable solution. It is quite popular to face these issues directly with non-supervised learning algorithms to train prey and predators. However, results got by most of these approaches still leave a great margin of improvement which should be exploited.In this paper we propose to start from a greedy strategy and extend and improve it by adding communication and machine learning. In this proposal, predator agents get a previous movement decision by using a greedy approach. Then, they focus on learning how to coordinate their own pre-decisions with the ones taken by other surrounding agents. Finally, they get a final decission trying to optimize their chase of the prey without colliding between them. For the learning step, a neuroevolution approach is used. The final results show improvements and leave room for open discussion.