Abstract/Summary

This thesis describes the development of a robotic control scheme with a novel ‘parallel’ design that combines direct feedback control with the simultaneous output of a dynamical forward model. Unlike existing predictive control schemes with a serial ‘sense-calculate-move’ structure where the model fully determines the robot’s behaviour, the parallel controller can adapt to overcome any feedback delay the system experiences without updating its parameters. This is thanks to replicating the key properties of anticipating synchronisation (AS), where a slave system (the robot) anticipates a similar master (a moving target) via delayed self-feedback. Since the robot and target possess very different dynamics, the output of the forward model is used to impose a suitable dynamical behaviour on the robot, while the direct feedback term simultaneously drives the robot itself to anticipate the target. This means that the forward model does not have to be related to the robot’s true dynamics so long as it represents a suitable AS slave system, and that any feedback delay will inevitably be opposed by a proportional degree of anticipation. The result is a highly robust and adaptable predictive controller that can be applied to a robot without requiring precise knowledge of its dynamics.