RamonaSrinivasan and his colleagues ran experiments on honeybees which indicated that they balance the magnitude of optic flow seen in their two eyes.  This will lead them to fly down the middle of a corridor for example.  This control law can be used for more general navigation as well in cluttered environments.  I implemented this control law and a few others on a small robot (Louie) which used optic flow for obstacle avoidance and to play a simple game of tag. This work was made possible using Ted Camus's optic flow algorithm.  I then extended these control laws for use in a much larger robot (Leslie Kaelbling's Ramona) using the Teleos AVP-100 vision system.
 
 

 
 


Agent Wandering: Perspective Agent wandering: Top
I've got a nice simulation of this on an SGI which allows a user to interact with the simulated robots.  Here are quicktime movies of an agent wandering around a simple environment from the top (1.4Mb) and from the agent's perspective (15.5Mb).

4 agents
Many agents running different control laws and with different fields of view can all interact  (4.5Mb).
 
 


TAG  This a QT movie of two agents playing tag  (1.7Mb), similar to that found in the papers.  These movies can be quite large and have a number of skipped frames.  Caveat observator.
 

 
 
 
 

Papers are available, including one that just came out in the journal Adaptive Behavior.  Reprints are available.

At least two other groups have implemented similar control laws:  David Coombs and Sandini's group at LIRA.


A preliminary attempt at folding these control laws into higher order navigation was present at Simulation of Adaptive Behavior in 1996 (paper).
 

Motor method There I presented three algorithms, one motor based where given a sequence you feel yourself moving by an opening or making a turn  (0.4Mb).
 
 
 

Perceptual method Another was more perceptually based in which you saw the choices available, made a random one and when you got to a cul-de-sac you reversed the sequence  (0.7Mb).
 
 
 

Salience Centroid Finally, I developed a more elaborate perceptual method based on some old models of hippocampal function  (1.6Mb).  For details, please read the paper.