Faculty Mentor

D. Michael Franklin, Ph.D.

Proposal Type

Oral Presentation

Start Date

3-11-2018 8:00 AM

End Date

3-11-2018 9:00 AM

Location

Nesbitt 2211

Abstract

Biologic predation is a complex interaction amongst sets of predators and prey operating within the same environment. There are many disparate factors for each member of each set to consider as they interact. Additionally, they each must seek food while avoiding other predators, meaning that they must prioritize their actions based on policies. eSense provides a powerful yet simplistic reinforcement learning algorithm that employs model-based behavior across multiple learning layers. These independent layers split the learning objectives across multiple layers, avoiding the learning-confusion common in many multi-agent systems. The new eSense 2.0 increases the number of layers and the amount of separation between agents so that the behaviors for each agent can be more highly customized and adds specific additional layers for behavior-only learning. In other words, each agent now has multiple layers to model each aspect of their behavior (e.g., obstacle avoidance, prey observation, prey seeking, etc.). This new abstraction of breaking out the various agent behaviors into multiple levels furthers speeds up the learning and clarifies the objectives the agent is considering. This significantly builds on the general goal of eSense (splitting out multiple agents into their own levels) because now the agent’s behaviors are also split out into multiple layers. The learning is now more expressive, faster, and less noisy. This presentation seeks to present this new multilevel learning system for multi-agent systems and confirm its performance through experimental results.

eSense2BioMimeticPredation.pdf (688 kB)
Full Research Paper

Share

COinS
 
Nov 3rd, 8:00 AM Nov 3rd, 9:00 AM

eSense 2.0: Modeling Biomimetic Predation with Multi-agent Multi-team Distributed Artificial Intelligence

Nesbitt 2211

Biologic predation is a complex interaction amongst sets of predators and prey operating within the same environment. There are many disparate factors for each member of each set to consider as they interact. Additionally, they each must seek food while avoiding other predators, meaning that they must prioritize their actions based on policies. eSense provides a powerful yet simplistic reinforcement learning algorithm that employs model-based behavior across multiple learning layers. These independent layers split the learning objectives across multiple layers, avoiding the learning-confusion common in many multi-agent systems. The new eSense 2.0 increases the number of layers and the amount of separation between agents so that the behaviors for each agent can be more highly customized and adds specific additional layers for behavior-only learning. In other words, each agent now has multiple layers to model each aspect of their behavior (e.g., obstacle avoidance, prey observation, prey seeking, etc.). This new abstraction of breaking out the various agent behaviors into multiple levels furthers speeds up the learning and clarifies the objectives the agent is considering. This significantly builds on the general goal of eSense (splitting out multiple agents into their own levels) because now the agent’s behaviors are also split out into multiple layers. The learning is now more expressive, faster, and less noisy. This presentation seeks to present this new multilevel learning system for multi-agent systems and confirm its performance through experimental results.