The Robot Sheepdog Project was a first investigation into the application of autonomous robots to control the behaviour of domestic animals. Three studentships were created to analyse and re-create the familiar sheepdog's task of gathering and fetching a flock to a specified goal position. The resulting theses present (1) a set of image processing techniques which can automatically generate models of flocking behaviour from video sequences \cite{sumpter:thesis}; (2) a study of the behavioural responses of the ducks when exposed to a robot-like vehicle \cite{henderson:thesis}; and (3) this thesis describes the development of an autonomous robot system that gathers a flock of ducks in a circular arena and manoeuvres them safely to a pre-determined goal position.
This work establishes a methodology for developing animal-interactive robots. An important feature of this methodology is that it enables the development of a machine that can usefully interact with an animal without using the animal in the design process.
Interacting with animals imposes strong constraints of real-time action, robustness and animal safety. A suitable arena, robot vehicle, control architecture and vision system are described.
It is not currently feasible to model the behaviour of any animal with anything like the precision available for a traditional engineering system. However, an animal-interactive robot must be robust with respect to the inevitable variations in behaviour between individual animals and even in the same animal over time. It is suggested that (a) animal-interactive robot controllers should exploit the underlying mechanisms of the subject animals' behaviour rather than the details of any particular animal or group, and therefore (b) a simple model of such an underlying mechanism can be used to aid the design a robot that will control the real animal system. Specifically, it was hypothesized that a robot controller that reliably gathers a simulated flock should also work when transferred to a real robot and flock of ducks.
A minimal generic flocking model is created and incorporated into a simulation of a the vehicle and arena. A simple robot controller is devised and demonstrated to work in simulation and transfer directly into the real world. A series of experiments are performed to assess the performance and reliability of the method. Consideration of these results leads to a second, simpler algorithm. The second method is tested in simulation and the real world and found to be more successful and more reliable than the first. A pair of no-robot control experiments gives benchmarks with which to compare the performance of the two methods.
Further experiments in simulation demonstrate the application of the flock control methods to on- robot sensing modalities; laser ranging and vision.
The thesis demonstrates the first robot interacting with animals to achieve a useful task. It is concluded that a minimal behavioural simulation can be a useful design tool for a real animal- interactive robot system.
RTV - last update August 99 - fixed links April 2004