Combining Rule-Based Knowledge with NEAT
Active from 2004 - 2006

Knowledge Based NEAT

Knowledge Based NEAT (KB-NEAT), combines the prescripted behavior contained within a FSM with the real-time adaptability of a neural network using NEAT. Using KB-NEAT, based on Knowledge Based Artificial Neural Networks (KBANN), developers can automatically convert a FSM into a network which exhibits the same policies then further evolve the network.

Movies (must have DivX Codec)
Experiment 1: A team of robots trying to chase a roving enemy around their spawn point.
Team spawned with random networks
Neural Network primed with a Finite State Machine using KB-NEAT to do the task.

Experiment 2: A team of robots must find its way around a turret's fire, to get close to the enemy. The primed network fails at this task and must learn to adapt.
Beginning of the experiment. The primed networks fails at the task
End of the experiment. The primed networks have adapted to run in a curved path around the enemy fire.

Computational Intelligence in Games 2006
Risto Miikkulainen, Bobby D. Bryant, Ryan Cornelius, Igor V. Karpov, Kenneth O. Stanley, and Chern Han Yong, In Computational Intelligence: Principles and Practice, Gary Y. Yen and David B. Fogel (Eds.), Piscataway, NJ 2006. IEEE Computational Intelligence Society.
Creating Intelligent Agents in Games 2006
Risto Miikkulainen, The Bridge (2006), pp. 5-13.
Improving Prescripted Agent Behavior with Neuroevolution 2005
Ryan Cornelius, Technical Report HR-05-01, Department of Computer Sciences, The University of Texas at Austin.