A rule-based lens modeling approach for assisted navigation training

This paper presents an assisted navigation training framework (ANTF) for users with severe motor disabilities that are only able to use human-machine interfaces (HMIs) characterized to provide a small set of commands, issued sparsely. With this framework we are pursuing the following goals: to train users to steer an assistive mobile robot (AMR) in an autonomous manner, to train users to operate HMIs, and to characterize the human user in both, steering the AMR and operating the HMI. The user characterization can latter be used to adapt the AMR navigation system to his/her steering and HMI operating capabilities. The training approach is based on judgment theory, more specifically on rule-based lens (RBL) paradigm, and is able to provide the characterization of individual performance in order to tailor training to the needs of each human user. Individual judgment performance is modeled using a genetic-based policy capturing (GBPC) technique characterized to infer non-compensatory judgment strategies from human decision data. The ANTF was designed to help users to navigate an AMR to a specific goal position in an indoor structured environment, and to characterize their performance, choosing the set of manoeuvres needed to perform certain navigation tasks. The ANTF is general enough to allow its usage with any HMI device providing a small set of sparsely issued commands (such as: scanning interface, brain-computer interface, etc.), allowing to choose the most suitable HMI device to the particular users´ disabilities. Three user models, at three different learning stages, using the RBL paradigm, are presented.