roland@quiche (Roland KUHN) (06/28/90)
To the readers of Comp.Robotics: I'm a graduate student in the Speech Recognition Group at McGill University. We're interested in designing a voice interface for a simulated robot, and we're having trouble tracking down a simulation that meets our needs. Ideally, it would interface with a graphic display on a Sun, so users can observe it moving around in response to their vocal commands. Most robot simulators seem to focus on low-level details. Because of the way people use language - very high-level - we're interested in something that can respond to commands like "Move to location <X,Y>". Obviously, this means we'll be working with something that does not accurately reflect the capabilities of today's robots - we have to simplify to some extent. On the other hand, we don't want to throw away realism completely. A typical dialogue might begin with the user giving commands to the robot. The robot asks for clarification of the commands and perhaps for information about the physical environment, then begins to carry out the commands. If it gets stuck, it may ask the user for "debugging" information. In real life, it may be several decades before such systems are practicable. Nevertheless, we're trying to design a natural-language interface (including a speech-recognition component) as a prototype. What we need is a small, high-level graphic simulation (preferably written in C and running on a Sun) that is not totally unrealistic, in that it gives us at least an idea of the kind of problems the naive user is likely to run into with a robot. We don't care much whether it models a robot arm or a mobile robot - anything with enough capabilities to give rise to an interesting "dialogue" will do. Any suggestions will be VERY gratefully received. Please e-mail them to roland@calvin.cs.mcgill.ca . Thanks! Roland Kuhn.