Mimic This: Societies of Curious Robots
In robotic systems, curious agents offer a way for developmental robots to select their own goals. Such robots have a range of potential applications, including support for tool use, fault tolerance and robot reconfigurability. Existing work with curious robots has focused on reward based learning approaches such as reinforcement learning. Such robots can learn by trial-and-error, but cannot draw on the experiences of other robots. This project will develop new models for curious robots that can learn to mimic interesting behaviours of other robots using supervised learning techniques. Models will be implemented and tested on the Lego Mindstorms NXT platform.
Description of Work:
- Review of relevant robotics and artificial intelligence literature
- Design computational models of curious supervised learning for continuous, real-world environments.
- Implement the models on the Lego Mindstorms NXT platform
- Evaluate the models using empirical metrics and/or case studies. This may include comparison to existing computational models of motivation or motivated learning.