Prof. Michael G. Dyer       HomePage

CS 263C Animat-Based Modeling

Animats are mobile/sensing animal/insect-like software agents embedded in simulated dynamic continuous environments. Emphasis is on modeling action selection and goal-oriented behavior via neurocontrollers and survival/adaptation of populations of animats via reinforcement learning and evolutionary programming. Animat-based tasks include: food foraging, mate finding, group predation, navigation, predator avoidance, cooperative nest construction, communication, and parenting. Course objective is to introduce students to issues that arise in modeling agents that must perform tasks while surviving and adapting to their environments. Issues include: smooth multi-modal sensory integration, motor coordination, learning from other agents (vicariously or via imitation), and achievement of sequences of high-level goals (e.g. for nest construction/repair) while still maintaining viability (e.g., eating when hungry; drinking when thirsty; avoiding predators, not getting lost, etc.).

Prerequistes: graduate standing in CS (or consent of instructor)

Topics Covered:

- Environments for Animat Evolution, Learning, and Development
- Representations of Animat Brains/Bodies
- Perceptual/Motor Modeling (simulated audition, olfaction, vision, coordination)
- Action Selection (reactivity, planning, attention, persistence of behavior, sequencing...)
- Learning (reinforcement, Hebbian, competitive, supervised, vicarious, imitative, ...)
- Development (topographic maps, neural and body growth and self-organization,...)
- Obstacle Avoidance and Navigation (trail laying/following, cognitive map use/formation, ...)
- Goals, Motivation, Emotions (goals via high-order connections)
- Evolution and Coevolution of Neurocontrollers (evolution of signalling and cooperation/competition,...)
- Situated Collective/Distributed Behavior (group hunting, nest construction, ...)
- Philosophical Issues (computational neuroethology, animats vs. robotics,...)
- Instrumentation and Model Evaluation

Hours per week: Lectures, 4 units (meets twice weekly)

Readings: taken from a variety of sources, with emphasis on articles from Simulation of Adaptive Behavior conferences (i.e., From Animals to Animats 1 - 7, MIT Press).

Grading: Based on: (1) course project and paper, (2) presentation of project, (3) presentation of a "favorite" paper. 

Offered: Very year, usually in the Fall.