Brian Gerkey
๐ค SpeakerAppearances Over Time
Podcast Appearances
dimensionality is very large.
So you're, and they vary over time.
You're reading from sensors, which might be cameras, they might be lasers, they might be inertial measurement units that tell you how you're flying around in the world and interpreting that data and then deciding what to do and then trying to, and then your action is not only like, I'm going to compute an answer and then write it down or display it to a screen.
The answer in a sense is like,
take action in the world.
So you're perceiving this world in this rich way, trying to understand what's happening out there.
And then the answer that you compute, instead of just being a number, it's a series of commands that you're sending to a robot and you're telling the motors what to do.
And you're telling them what to do with the understanding that they're not going to do exactly what you tell them to do, because there's always going to be some uncertainty in the
the in the action and so then you need to observe the world again close that loop and that it that kind of interaction with the physical world it's a pretty challenging problem so that's part of it is just dealing with that physical interaction another part that i think warrants special attention for robotics is that it is so interdisciplinary as a field so
If you want to build a robot, first of all, you need people who have the mechanical skills, which is totally not me.
I'm a software person.
I'm a computer scientist by training.
But you need the people who can build the physical body of the robot.
And then even once you've got that and you want to build the software on top of it, you need people who are experts at the underlying infrastructure.
You need people who are experts at interacting with databases.
You need people who are experts at building human machine interfaces because you're going to have a human actually interacting with this thing.
You need people who are experts in perception, in motion planning, in real-time control, in grasp planning.
And you need to provide a software platform that lets all those people collaborate together.
And frankly, that part of it, giving people the right tools so that they can focus on their specialty and then come up with a result and contribute it in a way that it plugs into a larger system without binding everybody up into trying to all work on the same code at the same time.
It's actually kind of tricky.