The remote control of robots is problematic if there is latency in the connection. We are most likely to control remote assets via the internet, with inconsistent latency, which may sometimes be very large. Clearly it is not appropriate to exercise direct control, as the robot may keep moving forward off the edge of the cliff if the stop command comes too late. The usual situation is supervisory control where instructions are emitted as tasks (e.g. move to a location, via a path that does not intersect the edge).
Our work extends the notion of supervisory control by:
i) the inclusion of consistent actions which appear to the user to be remotely controlled - remotely controlled actions have been done by others e.g. mimicking reflex actions to quickly avert damage, but often the implications for the supervisor interface is ignored. We add remote instinctive behaviour as used in our previous work in human robot/agent co-operation – the basic idea is that assets could appropriately pre-configure themselves for the next task and so on without explicit supervisor involvement; and
ii) we allow the remote system to observe the user - this requires high 'intelligence' in the remote system, but creates an analog of the interpersonal social intelligence we rely on with other human beings, or even dogs. That is, a human gesture, sudden head movement or facial reaction to an incorrect action may be enough to abort or at least slow down incorrect and potentially dangerous mis-actions.
Authors: Tom Gedeon, Dept Comp Sci, ANU
Event: SF08: Embodied Interaction in Mobile, Physical and Virtual Environments Workshop