A Case-Based Approach to Reactive Control for Autonomous Robots *

We propose a cause-based method of selecting behavior sets as an addition to traditional reactive robotic control systems. The new system (ACBARR -A Case BAsed Reactive Robotic system) provides more flexible performance in novel environments, as well as overcoming a standard "hard" problem for reactive systems, the box canyon. Additionally, ACBARR is designed in a manner which is intended to remain as close to pure reactive control as possible. Higher level reasoning and memory functions are intentionally kept to a minimum. As a result, the new reasoning does not significantly slow the system down from pure reactive speeds. not foresee. They are also incapable of taking advantage of navigational successes; even if a behavior has proven extremely successful, it is not used more than any other behavior in the system, nor is it allowed to alter to become even more successful. A third problem is that most reactive control systems do not utilize sets of behaviors; instead, they rely on a single type of behavior to guide the system. In order to achieve better robotic control, we feel that behavior sets need to be used along with adaptations of existing behaviors to new environments. One option is to have the system modify its current behavior based on immediate past experience. While useful, this is only a local response to the problem. A more global solution is to have the system select completely new sets of behaviors based on the current environment in which it finds itself.