Interactive robots need to be able to adapt quickly from user feedback, but getting lots of preferences takes a long time. It turns out that LLMs can serve as pretty high quality priors to speed up this process significantly. Check out Jaden's thread for more info! 👇

Hi! I'm gonna be presenting this at #ICLR2025 during the Thursday poster session (4/24; 3 p.m - 5:30 p.m, Hall 3 + Hall 2B #208). Come by if you want to talk about making ice cream!! (and also human-computer grounding, interacting with LMs, user models, etc.)

What makes data “good” for robot learning? We argue: it’s the data that drives closed-loop policy success! Introducing CUPID 💘, a method that curates demonstrations not by "quality" or appearance, but by how they influence policy behavior, using influence functions. (1/6)

Interested in data and robotics? Come to our CoRL workshop! We have a great lineup of speakers & a tutorial on data mixing. Submissions due Aug 22nd

🚨 We have extended the submission deadline for the workshop until Aug 29th! 🚨 Any completed or in progress work concerning data composition, curation, and interpretability in robotics are welcome!

We recently re-examined data retrieval methods through the lens of importance sampling, and found a simple modification that improves performance for several methods! We'll be presenting the paper at CoRL soon. Check out Amber's thread for more details!