AnimatLab: A Toolkit for Analysis and Simulation of the Neural Control of Behavior Ying Zhu Department of Computer Science Georgia State University SURA.

AnimatLab: A Toolkit for Analysis and Simulation of the Neural Control of Behavior

Ying ZhuDepartment of Computer Science

Georgia State University

SURA Cyberinfrastructure Workshop:

Life Sciences and the Grid

2

My Research Background

• Extensive experience on real-time 3D graphics, visual simulation, and medical visualization

• Recent projects– 3D visualization and simulation for

neuroscience – Collaborative virtual environment for

molecular modeling

3

Outline

• What is AnimatLab?

• Why build AnimatLab?

• Modeling and simulation of crayfish escape behavior

• The next generation AnimatLab and the Grid

• Summary

4

What is AnimatLab?

• A 3D computer graphics environment for neurobiologists to visualize and test computational models of neurons, neural circuits, sensors, and muscles, and their control of a model animal’s behavior in a physically realistic virtual world– Animat: artificial animals, including physical

robots and virtual simulations

5

AnimtLab Interface

6

System Architecture

7

Neural (Behavior) Editor

Behavioral Editor

Drag neuronsfrom the toolboxinto your network

Edit the propertiesof the selected

neuron

Draw connectionsbetween neuronsto add synapses

Create hierarchal neuralcircuits with multiple pages

Add nodes to connectneurons on one page to

neurons on another page

8

3D Body EditorData Chart Behavioral Editor

Body Plan Editor

Project Workspace

Property Grid

Simulation Controller

Simulation Window

9

Sensory Receptor

SensoryReceptive Field

Receptive Field Gain

ReceptiveField/Neuron

Pairs

AReceptiveFields

B C

1 2 3Neurons

One receptive fieldcan stimulate

multiple neurons

AReceptiveFields

1 2 3Neurons

Or one neuron can bestimulated by multiple

receptive fields

10

Simulation

11

Data Display

12

How does AnimatLab work?

13

Why build AnimatLab?

• A central goal of neuroscience is to understand how the nervous system is organized to control behavior

• This control must be dynamic and depend on a constant dialog between sensory input, including feedback, and motor commands

14

Why build AnimatLab?

• This important dynamic relationship between nervous function and behavior is poorly understood because of technical limitations to record neural activity in freely behaving animals

• Currently it is only possible to record from central neurons in restrained or anesthetized animals

15

Why build AnimatLab?

• AnimatLab can help formalize and evaluate hypotheses about the neural and physical mechanisms for dynamic control of behavior by simulating freely behaving animals

16

Related Works

• AnimatLab and other computational neuroscience tools (e.g. NEURON and GENESIS)

• AnimatLab and computational neuroethology

• AnimatLab and biorobots

17

Related Works

• Other computational simulationsof animal behavior exist

• But they were built for a specificanimal

• AnimatLab is a general purposetoolkit

18

Crayfish Escape Behavior

• The neural circuits of crayfish escape are among the best understood neural circuits in any animal, and for 60 years have provided a model for sensorimotor integration

19

Crayfish Escape Behavior

20

Create a 3D Crayfish Model

21

Simulation of Crayfish Escape

22

The Result• We were able to use AnimatLab to simulate the

fast abdominal flexion that evokes an upward directed movement of the model crayfish

• But the subsequent abdominal re-extension and swimming are ineffective

• The challenges: – Need more detailed neural model– Need more sophisticated muscle simulation– Need more realistic crayfish body parts– Some important circuit elements may not have

been identified

23

Next Generation AnimatLab

• A more powerful and extensible neural simulator

• A more extensible and transparent physics simulator architecture

• A more sophisticated muscle simulator

24

Next Generation AnimatLab

• An improved hydrodynamic simulator

• A better 3D body editor

• Optimization for new computer hardware

25

AnimtLab and the Grid

• Grid computing can – provide the ability to search through vast

parameter spaces such as various muscle parameters

– allow the user to evolve the neural network, the body of the organism, or both at the same time in order to meet some desired goal

26

AnimatLab and the Grid

• The grid would allow us to perform the search in a parrallel fashion on thousands of computers simultaniously.

• This vastly decreases the time it takes to perform such an evaluation.

27

AnimatLab and the Grid

• Grid services will be implemented as a plug-in for AnimatLab with four components– search algorithm– population generator– grid manager– visualization tools

28

Grid Computing at GSU

• GSU is deploying 1000 United Devices license across the campus

• We are working closely with Art Vandenberg’s group to take full advantage of this resource as well as SURAgrid

29

Summary

• We have been developing AnimatLab for 2 years

• Version 1.0 is expected to be released in the next six months for evaluation and user feedback

• Version 2.0 will be our focus for the next 3 – 5 years

• Interest among neuroscientists is high• AnimatLab will be a useful toolkit for

computational neuroscience

30

The Team• PI: Donald H. Edwards

– Professor of Biology – Director of GSU Brains & Behavior Program

• Co-PI: Ying Zhu (Computer Science) and Gennady Cymbalyuk (Physics)

• Collaborators: William Heitler (University of St. Andrews, UK) and Andrei Olifer (Emory University)

• PhD students: David Cofer, James Reid

31

Sponsors

Preliminary work has been funded by

• NIH P20-GM065762

• GSU Brains & Behavior Program

• A grant proposal was submitted to NSF Collaborative Research in Computational Neuroscience (CRCNS) in January 2006

32

Thank you!

• Questions?

[email protected] (Ying Zhu)

or

[email protected] (Don Edwards)