Designing and building behavior-based robotic systems are not an easy job. There is no general rule about how to build a complex system from a number of primitive behaviors. Moreover, designing interconnections between each behavior are not trivial as well. Some experiments need to be done to evaluate the performance of the system. Then, iteratively, fixing imperfect behaviors and adding new behaviors need to be done until the overall system shows good performance. In addition, learning process in behavior-based robot is also about training how to determine which behavioral actions are most appropriated for a given situation. The overall behavior system has to be designed before hand.
This research will propose the Automatic Behavior Programming, the way that the complex behavior system will be built up for a number of simple behavior models. All possible sensing inputs, statuses and actions are only need to define before hand. These features will be used to represent a huge possible number of Behavior Modules in this system. Linking together with five Behavior Interactions, a complex behavior system will be represented. By introducing Genetic Programming (GP) as learning process allows each behavior system to change processes, modules and interaction structures for selecting the most appropriate behaviors. In case to do that, structured behavior system will be represented in the structured programming system. A LISP-like language is used for representing each behavior system.
A few experiments with simulation robots will be presented.