Inverse Control for Inferring Intent in Novice Human-in-the-loop Iterative Learning

This article studies human-guided iterative learning control for output-trajectory tracking tasks when the controller does not have direct access to the intended output and needs to infer it from human demonstrations. Human intent (e. g., the desired trajectory) can be inferred by inverting known human-response models. However a challenge is that such human-response models are only available for limited types of controlled systems. The main contribution of this article is to enable inferring of human intent for general minimum-phase linear systems (which have a stable inverse) by using an inversion-based approach to specify that the apparent system perceived by the human user is one for which human-response models are known to exist. Then, the inferred intent (output-trajectory found by inverting the human-response model) is used to iteratively find precision output-tracking inputs. The proposed scheme is illustrated with a human-in-the-loop tracking experiment. Results show that the proposed inversion-based human-guided learning scheme allows the machine controller to learn from the human demonstrations (about 95% reduction in tracking error) with simultaneous reduction in the human control effort (about 96% reduction) as compared to the manual control by the human alone.