Self-Adjusting Finite State Machines: an approach to Real-Time Autonomous Behavior in Robots

In the Robotics industry, it is a frequent requirement that robots operate in real-time. The usual approach to this issue involves creating robots driven entirely by direct environmental input rather than complicated planning and decision-making AI. This approach means that the current state of the robot in relation to its environment exclusively determines the actions of the robot. In the simplest terms, this approach creates a Finite State Machine (FSM). Clearly, a standard FSM is completely pre-deterministic upon its creation. This is a drawback which immediately disallows the robot to cope with dynamic environments in an autonomous manner. This research suggests a solution to this problem, while still maintaining real-time performance of the FSM structure, through the development of a Self-Adjusting FSM (SAFSM). A SA-FSM is a FSM with an additional module which adds, removes, and adjusts specific states of its FSM structure. By adjusting its FSM the SA-FSM will have the basis for autonomous attributes. It will be capable of coping with drastic changes in its environment by making necessary fundamental adjustments to its behavior. Through this mechanism, the process of learning can be implemented. In this regard, only the inherent learning/inference algorithms the SA-FSM employs to adjust its FSM determine the complexity of the behavior produced by a SA-FSM based robot.