Journal article Open Access
Interfaces between human and a machine are at the forefront of research in human augmentation, assistance and rehabilitation.Initial acceptance of the interface relies on its intuitiveness, while continued use over a long period of time requires reliability.Co-adaptive interfaces have gained great popularity as a viable way to compensate for various sources of instability and to facilitate human operation. Nevertheless, correct functioning of the interface relies on user adaptation, a process that can be lengthy and cognitively demanding, and not enough effort has been devoted to address the mechanisms that enable the user to learn how to efficiently interact with the interface. This work proposes a mathematical framework for studying co-adaptation in body-machine interfaces that emphasizes the role of user’s learning in shaping the interaction with an adaptive interface. The framework formulates co-adaptation in a task-independent and model-free way assuming that the user and the interface co-adapt towards maximizing control efficiency. The generality of this novel framework can be exploited to simulate a variety of interaction scenarios, as knowledge of user intent or task goals is not required, it allows investigating the parameters leading to optimal co-adaptation dynamics and allows demonstrating the superiority of co-adaptation over user adaptation.