Peters Lab | Signal evolution

THE EVOLUTION OF ANIMAL MOTION SIGNALS: A LIZARD TAIL

Our goal is to evolve the temporal structure of tail-flicks, under the constraints imposed by environmental motion noise, energetic limits and predators. The general strategy is outlined in the adjacent figure. Briefly, we begin with a population of 100 random genomes initialized with random values (A) that define the temporal structure of tail-flicks (B-C). These signals are presented under varying environmental conditions to captive lizards (D). The fitness of each genome is determined using a function built around response probabilities and latencies (E). The distribution of fitness levels for the genome population (F) identifies the best performing genomes that will reproduce to the next generation (G). The descendent genomes undergo random crossovers and mutations (G) to define the second generation of genomes, and the cycle repeats.

We enlisted Ramana Kumar, an Advanced Science Student at the ANU, to develop a pilot simulation environment. The simulation environment was written in Ruby (a freely available programming language), can be accessed via a web browser, and makes use of a MySQL database to manage the information. Every detail of the simulations can be exported for further analysis, for example, to examine the change over generations of particular motion parameters (e.g., signal duration). This pilot work implemented all aspects of the simulation, with the exception of stimulus presentation to live animals, and allowed us to fine-tune our protocol.

1.	The upper level of the simulation environment lists each 'family' (or simulation experiment) and summarises the details of the simulation including mutation and crossover probabilities and the particular cost and benefit functions used (Screen 1).

Screen 1 - List of families and their properties

(Hide)

2.	The next level comprises a description of each genome in a given generation as well as response times (following presentation to live animals), fitness averaged across the animals that viewed the genome and the children of these genomes (Screen 2).

Screen 2 - Generation details including fitness & children

(Hide)

3.	A 'reproduction report' can be retrieved to track how children were determined as well as any pairings, crossovers and mutations (Screen 3).

Screen 3 - Reproduction report showing mutations, pairings & crossovers

(Hide)

The lowest level presents details of an individual genome including the bit string, flick information, fitness and reproduction information (NB: the pilot work was used to help refine the genome structure from coding 15 flicks and lead times for each flick to our current format of 10 flicks, flick speed and pause duration.) (Screen 4).

Screen 4 - Individual genome details

(Hide)

5.	After producing a new generation pressing the Animal Lscript button generates a script file (Screen 5)

Screen 5 - Script generated by simulator for animation software

(Hide)

6.	The script file is executed by the Lightwave animation software to create all animations (Screen 6).

Screen 6 - Animation example

(Hide)

7.	The final output is an uncompressed video file that will be presented to live lizards. The animations will be overlaid against wind blown plants - tail model and compositing details will differ to that shown in this illustration)(Screen 7).

Screen 7 - Rendered frame from an animation

(Hide)

(View an animation from an earlier experiment against a black background.)

Richard Peters & Jan Hemmi

RICHARD PETERS

Department of Zoology
La Trobe University
Bundoora VIC 3086
Australia

Tel. +61 3 9479 2234
Email. richard.peters@latrobe.edu.au