Dr. Kenneth Stanley, a Computer Scientist who studies Artificial Intelligence, is a young researcher with an inquisitive, creative mind. His primary area of study is Machine Learning, a branch of Artificial Intelligence (AI) that aims to teach computers and machines to learn and adapt in response to stimuli. Specifically, Ken's research focuses on the area of machine learning called neuroevolution, which means programming a machine to evolve small simulated brains called neural networks through an evolutionary process. Only the strongest and smartest neural networks survive.
Ken invented a computer algorithm called rtNEAT (Real-time NeuroEvolution of Augmenting Topologies) as well as the video game NERO (Neuro-Evolving Robotic Operations). NERO is unique in that its AI uses the rtNEAT technology. NERO was produced by a team of over 30 people, including programmers and computer artists. Both rtNEAT and NERO are generating significant attention in academic computer science circles, as well the gaming and simulation industries.
Ken is a recent Ph.D. graduate of the University of Texas Department of Computer Sciences (UTCS). In January 2006, he will take a position as professor at the University of Central Florida, where he hopes to continue his research in Machine Learning, neuroevolution, and AI.
While a Ph.D. candidate in 2003, Ken attended a conference on artificial intelligence and games. At the conference, the participants broke into small groups to discuss their ideas on how artificial intelligence might apply to video games in novel ways. He presented an idea for a game (NERO) based on rtNEAT to his group that was well received. The group selected the idea to be presented to the entire conference. If luck is about being prepared and being in the right place at the right time, Ken was a very lucky guy.
Unbeknownst to Ken, people from The Digital Media Collaboratory (DMC) Lab at the IC˛ Institute at UT Austin were
listening to the presentations and looking for a new project idea for the lab. The DMC Lab wanted to produce a video
game that could impact industries such as gaming, interactive digitital entertainment, and simulation. Because Ken's idea
intrigued them, they asked him if he would like to work with the lab to produce the game, noting that the DMC was willing
to fund its development. Video games are expensive and time-consuming projects that most graduate students could not dream
about doing alone, so Ken readily accepted their offer.
Middle top. Detail of sliders, which player uses to specify robot behavior.
Middle bottom. Detail of robots in battle that were trained to negotiate a wall.
Above: Left to right.
Aliza Gold, DMC Lab Coordinator and Project Manaager,
Michael Chrien, lead programmer
and UTCS student, and Dr. Ken Stanley, UTCS post doc.
Left: NERO game logo. The game has a futuristic setting.
The game was produced with the resources of the DMC, which is funded by prominent businessman, former Dean of the UT Business School, and UTCS affiliate George Kozmetsky. Another UTCS Ph.D. candidate, Bobby Bryant, helped design AI experiments during the production of NERO. Bobby and Ken's Ph.D. advisor, Professor Risto Mikkulainen, provided guidance and supervision. The DMC tapped many volunteer programmers and artists, mostly undergraduates, to assist on the 18-month project, eventually building a commercial-quality video game.
NERO allowed Ken to demonstrate the potential of neuroevolution in a novel way because the video game format is a convenient platform to showcase the research. Moreover, because of the versatility of rtNEAT, the project has generated commercial interest in other areas as well.
NERO (Neuro-Evolving Robotic Operations) could forever change the way non-player characters are programmed in the video gaming industry. NERO is unlike existing games in that players train their robotic soldiers to fight. Players can choose different behaviors for their robotic soldiers and devise different ways to teach them. Unlike characters in other games, the soldiers learn from experience, or die trying. They improve very quickly through real-time Darwinian evolution, and the best ones reproduce, just like in real life. Players, who act as drill sergeants in the game, train different teams of soldiers and pick and choose which trainees they want to use in each battle. For example, a player can mix and match different soldiers trained for different general skills such as aggression or avoidance, or a specific skill such as navigating a maze.
Special real-time technology was needed so the game could move fast and players could see the soldiers change quickly as the game progressed. rtNEAT (Real-time NeuroEvolution of Augmenting Topologies) is an enhancement of NEAT, a predecessor algorithm developed by Ken. rtNEAT allows the brains of robotic soldiers to evolve quickly enough to be fun for the player. During the course of the game, low performing brains die without reproducing, and high performing brains reproduce to create more even-higher-performing brains. The player can save the brains of high-performing soldiers for use in battle.
The long-term applications for rtNEAT and the NERO game are exciting and promising. In the future, the rtNEAT software could be used to produce smart automobiles and appliances, make military strategy decisions, or aid in education and job training. Although NERO is going to be freely available to the public, commercial games may follow. With the advent of rtNEAT software, one can assume the video gaming industry may never be the same.