A theory on autonomous driving algorithms A Swarm of Cars A theory on autonomous driving algorithms

Motivation Automobile accidents is one of the top 10 killers of people in the US More and more drivers join the roads each day—safety and efficiency is of primary concern Travel would become less “expensive”

Ideas Several theory’s have been suggested on how to accomplish this task The most prominent use a hierarchical scheme A SWARM system could also prove useful

Restrictions on Study Referring only to highway driving Vehicles using these methods have certain technologies: Speed sensors Road vs. off-road sensors Acceleration rates and stopping speed must be available

Needs System must exceed today’s safety System must be comfortable Fewer collisions System must be comfortable A roller-coaster type ride would not be acceptable System must be adaptable Replacing every vehicle on the road is never an option

The Answer from the Sea Schools of Fish exhibit all the needs of our system naturally Very, very, very, very rarely collide Move in smooth motions when not under attack Schools vary in size from 10’s to 1,000,000’s

A little background… Fish sense obstacles and other fish around them with the Lateral Line Sense Works similar to our ear We must mimic this with our technology

But wait… Fish schools have already been modeled by computers Used mostly in Computer Graphics Fish are modeled by using the Flocking behavioral model My work has been on adapting this model to fit onto a freeway

Three behaviors There are three behaviors that a flocking SWARM unit exhibits Separation The tendency of a unit to move away from others Alignment The Tendency of a unit to point in the same direction as others Cohesion The Tendency of a unit to move towards others

Separation Simple function: This doesn’t quite work for our purposes V = - mS(1/D)/N V is the placement vector D is the distance vector between the unit and the obstacle N is the number of obstacles m is a multiplier This doesn’t quite work for our purposes

Separation Separation zone should not be static Should be related to the stopping distance at a given speed X = -V^2/2a + b for directly in front of the vehicle Smaller for sides, area behind is irrelevent X = sin(T)*-V^2/2a + b for 0 < T < 180, X=b otherwise

Separation Vehicles cannot turn around Therefore positioning must be relative

Alignment Alignment zone should be similarly shaped Larger Alignment Algorithm: Vehicle angle is the average angle of all the units in the alignment zone Again, doesn’t quite work for our purposes Need to ignore vehicles traveling in the opposite direction Need Time Delay

Cohesion Works in opposition to Separation algorithm Should be the largest zone, and actively searching for new members to flock with Need other members to share information Movement vector is opposite of Separation: V = mSD/N

Emergent behavior Every member of the flock “sees” what the members at the front “see” Members move in unison Members will avoid obstacles in the same motion

Comparison Vs. Hierarchical Network AHS Pros: Deployable as an “option” No single point of failure No tracking movements Cons: No effective way to avoid congestion

Questions?