Experience-based Learning Mechanism with a Concept of Vigilance

Cognitive Neuroscience studies have identified an early warning system in the human brain that can avoid to make past mistakes again. They have shown how the brain remembers details about past dangers [1]. An activity was found in the Anterior Cingulate Cortex (ACC) after making mistakes [2]. This cortex area works as an early warning system that adjusts its behaviour to avoid dangerous situations. It responds not only to the sources of errors (external error feedback), but also to the earliest sources of error information available (internal error detection) [3]. It becomes active in proportion to the occurrence likelihood of an error [4]. Therefore, it can learn to identify situations where humans may make mistakes, and then help avoid such situations to occur [2]. It learns to predict error likelihood even for situations where no error occurs previously. Through the observation of particular areas located in cerebral cortex that has been shown to be responsible for cognitive control. Neuropsychological studies demonstrated a switching in human learning strategies around the age of twelve years. This switching, goes from learning with positive feedback to learning with negative feedback -probably comes from the combination of brain maturing and experience [5]. we have produced an early warning mechanism that can help avoid repeating past errors in the generation of bipedal motion patterns for a humanoid robot to achieve robust walking. The objectives of this learning mechanism is to adapt parameters of a low-level controller. In detecting its domain of viability, which increases adaptation to external perturbations [6][7]. We specified by the state space “V” of those intrinsic parameters. The mechanism must be able to learn from negative feedback (failure) and positive feedback (success). Therefore, it must have experience with success and other with failure within the state space “V”. As each vector “v” from “V” leads to either success or failure, the mechanism will evaluate whether this vector belongs to the success domain or to the failure domain. The decision mechanism (“go”, “nogo”) works as an early warning system similar to that in ACC [2]. Psychological studies suggest that some people are more tolerant to risk than others who are Biologically-inspired neural controller based on adaptive reward learning