Learning by Demonstration for Co-Operative Navigation with Assistive Mobility Devices

Publication Type:
Conference Proceeding
Citation:
2015, pp. ? - ? (7)
Issue Date:
2015-12-02
Full metadata record
Files in This Item:
Filename Description Size
acra15_published.pdfPublished version853.32 kB
Unknown
This work proposes a learning by demonstration framework for intuitive navigational adaptation of human-robot interactive mobile systems. Co-navigation algorithms for mobile robots tend to be highly parameterised with variables that may be difficult to manually configure to an individual user's desired behaviours by non-technical personnel, e.g. carers and therapists overseing activities with would-be intelligent power mobility devices. The proposed framework automatically learns suitable joystick inputs for safe handling of the platform from a healthy user aware of a desired subset of behaviours (generic collision avoidance, wall-following and forward/reverse alignment manoeuvres) through performance of a small set of elementary training exercises, without the need and risk associated to trial-and-error variable tuning. The paper compares the semi-autonomous capability of the proposed learning scheme with the popular Vector Field Histogram local planner in a corridor navigation task, showing its ability to safely generalise to different environments despite the simplicity of the training demonstrations.
Please use this identifier to cite or link to this item: