katherineliu.bsky.social
Senior Research Scientist at the Toyota Research Institute, previously Robotics PhD at CSAIL/MIT. Excited about embodied intelligence and the intersection of CV/Planning/ML. Opinions my own. Give me all the robots.
37 posts
1,945 followers
108 following
Regular Contributor
Active Commenter
comment in response to
post
Thanks for the suggestion and your contribution to the community! What year did you organize?
comment in response to
post
I personally tuned in (from a laptop -- the room was very crowded) to the third debate ("Generative AI will make a lot of traditional robotics approaches obsolete.") and highly recommend it!
comment in response to
post
Thanks for the context!
comment in response to
post
Lots of protein-related answers in this thread right now! I’m an outsider to the field — do the new diffusion-based methods represent a step change in performance, or is it more incremental?
comment in response to
post
I hadn’t read any weather prediction ones yet — thank you for the link!
comment in response to
post
Multi-agent SLAM is definitely a thing -- this should be a good recent work to start with web.mit.edu/sparklab/202...
And there's also a repo here
github.com/MIT-SPARK/Ki...
comment in response to
post
A great example!
comment in response to
post
Conditioning via reachability is really interesting! Adding to my to read pile.
comment in response to
post
(By which I also simultaneously mean please keep it coming 😂)
comment in response to
post
I really like this direction, no propaganda necessary!
comment in response to
post
My colleagues at TRI also just won an Outstanding Paper Award at CoRL for their work on diffusing physical parameters for control. Drifting Supras are close to peak cool. One Model to Drift Them All:
openreview.net/pdf?id=0gDba...
comment in response to
post
Here are a couple I’ve been involved with/seen: We presented work this year at IROS showing that you can do pose estimation by diffusing NOCS maps, naturally handling difficult practical problems like symmetry.
woven-planet.github.io/DiffusionNOCS/
(led by Taku and Sergey, who are not here yet)
comment in response to
post
Thank you!
comment in response to
post
✋Would love to be added!
comment in response to
post
The name of the list is obviously still aspirational as I wait for ICRA/CoRL/RSS/etc 👀
comment in response to
post
✋
comment in response to
post
@cpaxton.bsky.social manages the Robotics and AI one! You can see who created each list if you click on them; I think it’s also usually the first person on the list.
comment in response to
post
And finally, a call to action: Please share your robots! (Ratings optional)
comment in response to
post
The hacker/maker culture is also great strength of the robotics community to take advantage of; I’m really excited to see manipulators going through a similar democratization that quadrotors did back in the day.
comment in response to
post
If I could go back and talk to my younger self, I’d point out that robotics isn’t solved yet (true then and now), and the great thing about research is that we’re collectively exploring. Most of the projects listed in this thread were team efforts, with of folks with diverse technical backgrounds.
comment in response to
post
On a more serious note, putting together this non-serious thread was a great reminder of how incredibly fun robotics is. My advice to budding roboticists is to jump in! I wasted time when I was younger worrying about expressing interest in things I didn’t have the perfect credentials for.
comment in response to
post
Franka Panda, c. 2023 - present. At TRI, I’ve been really excited to explore robot arms. 14/10, can actually interact with the physical world instead of just trying not to crash into things.
franka.de
comment in response to
post
MIT RACECAR, c. 2019. We adopted this platform for some navigation research. 13/10, worked even though for our ICRA experiments we mounted the sensors to a cardboard box. (Speaking of which, I was sad to see that the Intel T265 was discontinued -- that thing was solid.)
racecar.mit.edu
comment in response to
post
Large Quadrotors, c. 2015
After joining the Robust Robotics Group at CSAIL/MIT (led by Nick Roy), I worked on quadrotors for the DARPA FLA and NASA SaRUC projects. 14/10, experienced the highest of highs (flying super fast!) and lowest of lows (crashing!). www.youtube.com/watch?v=2hRN...
comment in response to
post
Small Quadrotors, c. 2015. In my first year of grad school, I took a super fun class that involved writing controllers in MATLAB and uploading them to these small quadrotors. 12/10, survived being crashed into my ceiling as I figured out how to tune gains. fast.scripts.mit.edu/dronecontrol...
comment in response to
post
EyeRover, c. 2014. An inverted pendulum robot I worked on while an intern at BrainCorp. Super cute, could kind of just roll in a very fun way around but still 10/10.
www.technologyreview.com/2014/09/22/1...
comment in response to
post
TurtleBots, c. 2014. My first time using Linux, in Jorge Cortes’s lab. 13/10, great little robot that contributed to science just by toting around a laptop and spinning around.
comment in response to
post
First up, Falco, circa 2013. We constructed FALCO with wet layups in a basement at UCSD for the AUVSI sUAS competition. Falco was tasked with truly wild tasks like detecting all the alphabetic letters in a field to decode a message and flying over a router to receive a message. 12/10, flew high.
comment in response to
post
From what I can tell, no -- although it definitely seems like it would be a really useful feature!