2005 Computerworld Honors Program

	Virtual Football Trainer University of Michigan Ann Arbor, MI United States Year: 2003 Status: Laureate Category: Education & Academia Nominating Company: EMC
	Immersive virtual reality simulations train football players to recognize quickly critical play situations and to react instantly to opponent's moves, inspiring a spin-off company that will market the technology to professional teams.
	In American football, players must not only posses athletic skills. They also need to be trained for the correct visual perception of an evolving play situation and for the fast reaction to the movements of other players on either team. These intellectual skills are often difficult to master and require extensive field practice and game experience. Currently, players need to memorize hundreds of schematic playcharts and translate these two-dimensional drawings mentally into the situation on the field, where they are confronted by fast moving, large-size, three-dimensional athletes. The technology of immersive virtual reality offers new ways for the training of these intellectual skills. Using a virtual reality CAVE, a trainee can be placed at any location of a virtual football field and can be surrounded by three-dimensional, full-size players that are computer generated and animated for the simulation of a given play. A CAVE is a room-sized cube (typically 10x10x10 feet) consisting of three walls and a floor. These four surfaces serve as projection screens for computer generated stereo images. The users entering the CAVE wear lightweight LCD shutter glasses that provide three-dimensional stereo viewing. The effect is so compelling that the walls and corners of the CAVE are mentally blocked out by the human brain. The floor projection allows three-dimensional objects (like virtual football players) to appear inside the CAVE room and confront the user in a convincing way. Through the surrounding walls, the CAVE provides ultimate immersion and an extreme wide field of view. Peripheral vision is well supported and important for orientation and the perception of movements that occur in the periphery. These aspects of the CAVE are instrumental for the Virtual Football Trainer and allow for the perception of a given play situation, the correct estimation of distances, the awareness of the location of players on either team, or the visual communication with the (virtual) coach on the sideline. The animation of 22 virtual players that move according to a given play pattern requires animation experts and endless hours of hard work. Since football players need to be trained for hundreds of plays and play variations, the creation of these animations becomes an insurmountable task. We developed a new approach as a solution to this problem. Starting with conventional tools, the trainer (usually a coach) uses a "Chart Editor" on a laptop computer to outline a new play in two dimensions with symbols and arrows indicating the player's positions and moving patterns. This chart serves as input to an artificial intelligence algorithm that automatically creates the three-dimensional animations for use in the CAVE. This automatic process is at the heart of the Virtual Football Trainer. It makes the application feasible and allows to create new plays in minutes and play variations in seconds. It also provides a new tool for more serious applications. Training applications that involve virtual humans like combat simulations, training of squad teams for dangerous missions, police deployment, or simulation of accidents that involve people have the same problem. The virtual humans in these applications need to be animated according to a given simulation script. Our artificial intelligence algorithm can created these animations automatically from two-dimensional charts. The Virtual Football Trainer is not the product of a single person. A team of dedicated students and staff contributed to this successful development: Thana Chirapiwat, Mira Dontcheva, Bruno Janz, Lars Jensen, Denis Kalkofen, Joshua Levi, Joon Young Park, Diganta Saha, and Lars Schumann.
	The Virtual Football Trainer is presently a prototype application that is ready for conversion into a commercial product. The prototype has been tested intensively by football professionals. The benefits have been best summarized by Tom Brady, former Michigan quarterback, as follows: - Young players can go through countless repetitions of their own team's plays and learn the hundreds of formations, motions and variations to an offensive and defensive scheme. - An opponent's tendencies can be programmed and the simulation can be used to prepare quarterbacks and linebackers for anticipated schemes in an upcoming game. - Players can spend additional time in the CAVE to review simulated defensive maneuvers, such as opposing blitzes versus pass protections, over and over again. - Coaches can use the Virtual Football Trainer as a priceless tool in the off-season to train players. They can also visualize new plays to install in their own game plans. To make the application available to users without access to an expensive CAVE system, we also developed a version that runs on the World Wide Web and uses the VRML (Virtual Reality Modeling Language) Standard. We envision for the future that "Monday Morning Quarterbacks" can download the virtual version of a completed play, place themselves in the position of the quarterback, and try their own version of this play. The long-range benefit of this development is the potential use of our artificial intelligence algorithms in other areas of national importance, like combat simulations, training of squad teams and police forces, or the simulation of accidents involving humans. As in football, visual perception and quick decision making can be trained using the principles and technologies developed in our application.
	The development of the Virtual Football Trainer used a broad spectrum of information technologies, including three-dimensional computer modeling, skeleton-based animation of virtual humans, real-time computer graphics rendering of 3D environments, immersive virtual reality techniques, artificial intelligence, VRML on the World Wide Web, and more. Without these technologies, the application would not have been possible. We see our major contribution in the overall design of this training application and, specifically, in the development of artificial intelligence algorithms that convert two-dimensional sketches into fully animated three-dimensional scenarios.
	The original idea for the Virtual Football Trainer was inspired by the Lloyd Carr, head football coach at the University of Michigan. After finding initial funding through a gift to our university, designing the entire system and making it work was a tremendous task. It required original thinking and the solution of endless problems. A breakthrough came with the idea of automatically creating the three-dimensional animations. According to our knowledge, this approach has never been implemented anywhere. The University of Michigan has filed a patent application for this aspect of the application.
	As soon as the Virtual Football Trainer project was publicized on our Web site (www-VRL.umich.edu/project/football/), we received enthusiastic interest from football fans and professionals. The media started to report on this project and we could hardly handle the requests for demonstrations. This is understandable considering the popularity of American football. But we also received inquiries from other countries asking for possible conversion of the application to European soccer, Canadian ice hockey, and other sports. Most important was the success in our search for funding that would allow us to convert our current prototype into a commercial product for use by the NFL, by college football teams, as well as in high school football. Fortunately, several venture capital investors visited us. As of now, a new company has been founded by football professionals with the goal to bring our application to the market place. Negotiations with the University of Michigan regarding a license agreement are in their final stages.
	The biggest challenge of the project was the automatic creation of an animated play from a two-dimensional playchart. The movements of the virtual players needed to be created on a dense time grid. Our algorithm recognizes if a player is walking, running, stumbling, tackling, and throwing or catching the ball. Using collision detection methods, interactions between players that come close to each other are handled properly. The human visual system is very sensitive in recognizing movements of other humans and can easily distinguish between realistic and artificial animations. The best animations are based on motion-capture data, i.e., data that have been created by tracking the movements of a real human. We did not have the resources for creating motion-capture files, but will hopefully be able to do so in the future.