An approach to Artificial Intelligence and robotics promoted by Professor Rodney Brooks of the MIT Artificial Intelligence Lab. Traditional approaches to AI and robotics have generally followed the SMPA or sense, model, plan, act paradigm. In that paradigm, the robot first uses its sensors to see what the world looks like, then builds up an internal model of the world inside its "brain", builds a plan based on what it is supposed to be doing, and then sends commands to its actuators that actually cause it to move based on the plan it just computed. Of course, this approach to robotics has only met with limited success, as it requires vast quantities of computing power and memory, and has been to this date totally unsuccessful in producing a fully autonomous robot. In contrast, behavior-based robotics totally eschews the idea of explicit knowledge representation and plan formulation but instead endows the robot with behaviors that based on sensory input are somehow wired to the actuators that controls the robot's actions, just like reflexes in actual biological organisms. Instead of modelling the world, as the SMPA approach requires, this paradigm would have the robot "let the world be its own best model" as Brooks puts it. This approach, on the other hand, has been far more successful in making a workable autonomous robot, and requires far less computing power than the traditional methods.

The most important method for implementing this paradigm was described by Brooks in his seminal 1985 paper: "A Robust Layered Control System for a Mobile Robot". The paper describes the subsumption architecture, where a hierarchy of asynchronous behaviors is programmed into the robot, which read the robot's sensors and send prospective commands to the actuators. A subsumption module then prioritizes these behaviors and controls which actuator outputs of the robot actually get sent to the robot's physical actuators and control the actions of the robot. Higher level behaviors have the ability to suppress the prospective outputs of lower level behaviors.

One example of this architecture in action would be in a mobile robot that capable of sticking to a wall. We would place one sonar sensor on the front, left, and right sides of the robot, readings of which would give the distance of the robot from any obstacles in those directions. We would then program two levels of behavior into the robot, both of which read the sonar sensors and try to make the robot perform a particular action. The lower level behavior attempts to make the robot find a wall; it continuously reads the sensors to see where the closest wall or other obstacle is and move towards it, turn towards the closest wall. The higher level behavior tries to keep the robot from coming too close and crashing into the walls; steer the robot to the right if there's a wall too close either in front of us or to the left, or turn left if there's a wall too close to the right.

It's easy to see that this robot will try to stay near a wall it can find, as if it feels it's getting too close to a wall, the higher level behavior will activate and cause it to turn away from that wall. On the other hand, the lower level behavior will cause it to move towards the walls. The interaction of these behaviors and the environment result in the robot moving alongside a wall, not coming closer than any minimum distance and not moving further than a maximum distance. If the wall is part of a maze that has been properly constucted (i.e. without any paths that cross), the robot will follow the maze. This illustrates a central tenet in behavior-based robotics research: "intelligent behavior" is an emergent property of systems. We did not program the robot to traverse mazes, it never planned to traverse the maze, yet it showed the behavior of being able to traverse a maze because of the fusion between the behaviors and the environment the robot was in. This is known as "behavior fusion", as opposed to the sensor fusion sought after by SMPA practitioners.

Other interesting papers by Brooks and his associates on this topic of robotics and artificial intelligence are "Intelligence Without Reason", which is basically Brooks' manifesto for his approach to AI and "Fast, Cheap, and Out of Control".

Log in or registerto write something here or to contact authors.