"RoboSoccer in the Cave"

RoboCup (Robot Soccer) is a standard problem from Artificial Intelligence. Its goal is to let teams of cooperating autonomous agents play a soccer match, using either real robots or simulated players. We constructed a VR environment in which humans in CAVEs at different geographic locations can play along with a running RoboCup simulation in a natural way. A central role is played by the so called Soccer Server, which keeps track of the state of the game and provides the players with information on the game. The players are individual processes that can request state information from the server and autonomously compute their behavior. The server also enforces the rules of RoboCup and ignores invalid commands from the players.

Our RoboCup VR system uses the unmodified existing server software. The players communicate with the server by sending soccer commands. The commands are expressed in a simple language, consisting of accelerations, turns, and kicks. The server discretizes time into slots and only the last command of a player within a time slot is executed. Also, the kick command requires the player to be close to the ball. We run such a system on a Myrinet cluster computer.

The CAVE program allows the user to be immersed in the game and to interact with it. We implemented a proxy which uses the same information and communication as the existing 2D visualization. The data set described in our CAVEStudy configuration file consists mainly of the player positions and the ball position. From successive states of the game, the visualization system computes several quantities such as direction, velocity and acceleration of the players. We built a virtual stadium and a parameterized soccer player whose movements are interpolated between three different modes: standing still, walking, and running. We developed software to track the behavior of a human in the CAVE. One tracker is connected to the viewing glasses and monitors positional changes of the human player inside the CAVE. The second tracker is connected to the wand, which is used for global movements over the soccer field. The third tracker is attached to the foot of the human player and is used to recognize a kick. We convert tracker changes into soccer commands and transmit these to the server. Finally, we coupled two CAVEs located in Amsterdam and Stockholm. The two CAVEs are connected to the same proxy, so the two humans participate in the same game.

The most difficult problem in realizing a virtual RoboCup system is caused by the latency of the simulation program. If the human player moves over the virtual soccer field, these moves happen almost instantaneously for the human. In contrast, the soccer server will require some time to process the change of position. Also, the wide-area (Internet) connection causes a substantial delay. This problem is a typical example of how a delay introduced by a simulation program can harm a natural and real-time interaction. We are currently developing accurate and low-bandwidth algorithms for the navigation (walking across the soccer field) and the interaction (kicking detection) that generate commands to the server.

The first implementation of this system was done without CAVEStudy. We had to program the communication between the different components, which is tedious and error prone. Using CAVEStudy, the communication is automatically generated. More generally, the case study with RoboCup shows the applicability of CAVEStudy on the large class of agent/server systems.