Bob Iannucci: Virtual image processing environment for research


We have an object that was motivated by all
of this training of drones, called VIPER, which is the Virtual Image Processing Environment
for Research. Let’s say we’re building a drone control
system where the objective is for the drone to learn not to crash into things. We can let the drone crash 10,000 times so
that it learns how not to crash. Doing that in the real physical world has
some negative consequences: you end up with 10,000 drone carcasses. If you’ve seen the movie The Matrix, you
know that Neo goes into this dojo and learns how to be a kung-fu artist. But the training was sort of simulated; he
woke up and he says, “I know how to do kung-fu.” Imagine doing the same thing for apps. If we could take the structural component
of an app and provide it with virtual training in an environment that’s photo-realistic—where
the physics in the environment is like the physics that we know of—and we can do it
faster than real time, then we have the ability to train this structural component to do light
pole recognition, to inspect factories, or to look for traffic jams, or what have you. And that’s the idea of VIPER. We’ve taken a drone control program simulator
and a photo-realistic image rendering—a game engine actually—and we put them together
into a system that allows us to fly the drone in simulation, see what the drone sees in
real time (or even faster than real time) so we can have a virtual camera that sits
on the virtual drone in the virtual environment that feeds a virtual video to this program
that needs to be trained. And we can do the same kind of training in
this virtual world that I just described doing in the real physical world. So we’re looking for ways to do exactly
that. By doing the drone research we’re learning
about how we can distribute computation but put a significant amount of intelligence into
the drone and minimize its communication (e.g. “Please tell me when you see this object.”) In the future we might do that with a sensor
that’s in the city. Same programming environment and we take the
drone idea today, trickle down technology finds its way into the smart city tomorrow.

Leave a Reply

Your email address will not be published. Required fields are marked *