Abstract: This paper explores an application of an adaptive learning mechanism for robots based on the natural immune system. Most of the research carried out so far are based either on the innate or adaptive characteristics of the immune system, we present a combination of these to achieve behavior arbitration wherein a robot learns to detect vulnerable areas of a track and adapts to the required speed over such portions. The test bed comprises of two Lego robots deployed simultaneously on two predefined near concentric tracks with the outer robot capable of helping the inner one when it misaligns. The helper robot works in a damage-control mode by realigning itself to guide the other robot back onto its track. The panic-stricken robot records the conditions under which it was misaligned and learns to detect and adapt under similar conditions thereby making the overall system immune to such failures.
Abstract: In this work, we suggested a new approach for the
control of a mobile robot capable of being a building block of an
intelligent agent. This approach includes obstacle avoidance and goal
tracking implemented as two different sliding mode controllers. A
geometry based behavior arbitration is proposed for fusing the two
outputs. Proposed structure is tested on simulations and real robot.
Results have confirmed the high performance of the method.