Immersion for analyzing complex software.

Scanning the millions of lines of a computer program's instructions in order to find inconsistencies and other problems is a time-consuming, error-prone task -- even when the process is automated.

Researchers are now developing a system that generates a three-dimensional representation of a computer program, letting programmers visualize, experiment with, and modify the software. In effect, the technique immerses a person in the software, says Thomas P. Caudell of the University of New Mexico in Albuquerque. "It's like being inside a brain."

At a high-integrity software conference last week in Albuquerque, Caudell described a prototype virtual reality system, running on a supercomputer, for analyzing complex computer systems.

As a first step, he and his coworkers have used their approach to track what happens within a complicated computer program -- an artificial neural network -- that drives a simple robot resembling a roller skate. Called Encephalon, the neural network incorporates modules for processing sensor data, controlling the robot's actions, and performing other functions.

In the virtual reality environment, each software component is represented by a three-dimensional block that flashes colors or makes sounds as the component functions. Paths along which data pass appear as links between blocks. Wearing special goggles, the researcher sees the robot sensor data as if they were projected on the walls of a computer-generated virtual room. The linked blocks float in the middle.

Navigating around the room, the user can modify the software, set or adjust system parameters, monitor information flow, visualize intermediate computational results, view the raw input and output data, evaluate the system's overall performance, and observe the resulting behavior of the robot.

The team's current focus on artificial neural networks is only one possible application. Caudell says. Similar schemes could be used for simulating industrial process control, visualizing enormous databases to facilitate pattern recognition, and improving engineering and software design.

"A large proportion of all computer problems is attributable to the initial, informal, subjective phase of conceptualizing how a system should or should not behave," says Larry J. Dalton of Sandia National Laboratories in Albuquerque. Virtual reality potentially offers a way of exploring such behavior.

At present, however, many research questions remain unanswered, particularly the issue of how best to represent abstract entities in a three-dimensional environment. "We're trying to extend natural human capabilities of seeing patterns and organizing knowledge," Caudell says, "but we need to understand more about how people think and how the brain works."