This project is focused on creating a software framework that can be used, in a variety of domains, to imitate the behaviour of a software agent. Our approach uses case-based reasoning to successfully imitate software agents with minimal knowledge of the task being imitated.
Our approach involves watching a teacher perform a task, such as playing soccer, and memorizing how the teacher reacts (what actions the teacher performs) given the location of objects in its field of vision (what it can currently see). The imitator will therefore only need a sensory system similar to that of the teacher (so it can “see” the same things) and the ability to physically perform the same actions. Our software can then be used to process the sensory data and perform the appropriate action (the action the teacher would have performed).
- Robotic control – Imitation of a robotic control model that follows the Discrete Event System Specification (DEVS) modeling formalism.
- Transfer to a robot – Transferring behaviour, using imitation learning, from a RoboCup agent to a robot.
- Transfer from a robot – This time the RoboCup agent is learning from the robot.