RobotLocomotion / drake

Model-based design and verification for robotics.
https://drake.mit.edu
Other
3.36k stars 1.27k forks source link

Where do I ask this type of question? Can/Will you answer it for me? #20999

Closed josh-ramer closed 9 months ago

josh-ramer commented 9 months ago

What happened?

StackOverflow only allows programming related questions, emailing famous MIT professors seems very likely to be firing into the abyss, so, is there some way that the user community discusses the applications of Drake. I want to know specifically, which companies are using Drake & how are they using it, but I can't find the forum to ask such a question & I'd love to either be pointed to this sort of information or directed to the appropriate channel. It doesn't need to be exhaustive, but I'd love to see a few examples outside of research in which Drake is actively in use & learn about that experience. I did see the 2021 TRI article in which Russ discusses this briefly, but I'd love the 2024 update.

I filed this as a bug because it doesn't fit anywhere & that's buggy to me, also the furious defenders of StackOverflow didn't let attempt #1 last more than a few minutes.

Version

No response

What operating system are you using?

No response

What installation option are you using?

No response

Relevant log output

No response

RussTedrake commented 9 months ago

Here's a quick response from the abyss...

Many companies and research labs use Drake. TRI's leadership is clear. Amazon Robotics and Dexai are two great examples that have been very public about it and have been actively contributing back to the codebase.

In terms of physics simulation, Drake plays a niche role as a fully featured dynamics engine that has targeted accurate simulation and small sim2real gaps at fidelities that are approachable at 1x or greater real-time simulation speed. For instance, we've seen a number of applications where people use a different GPU-based simulators to train a policy using reinforcement learning from scratch, but then fine-tune in Drake to close the sim2real gap. As we are entering the age of foundation models, the need for higher fidelity sim has been growing.

As the medium article you've found explains, Drake is much more than just a physics engine. The systems framework has proven to be a foundation that large companies can build their broader sim infrastructure around. And the code maturity, aggressive license tracking, and long deprecation timelines makes it possible to build on it in production workflows.

Finally, the connection to the optimization libraries enables optimization-based robotics research for a lot of groups. And you can find mature algorithms for inverse kinematics and motion planning (including some of our latest and greatest based on Graphs of Convex Sets).