UW News put out a Q&A about our recent work on Variational Preference Learning, a technique for personalizing Reinforcement Learning from Human Feedback (RLHF) washington.edu/news/2024/12...
Happy holidays from UW Robotics!
❗️Problem: Robots navigating with no prior maps relying only on local sensors have a limited mapping range (due to sparse/noisy depth) causing myopic decisions.
🧵2/6
Thank you!
Yea we also went through a lot of the papers that tried to do long range perception for the LAGR project.
Really cool to take inspiration from works almost 20 years old but still very relevant :)
Long Range Navigator (LRN) đź§â€” an approach to extend planning horizons for off-road navigation given no prior maps. Using vision LRN makes longer-range decisions by spotting navigation frontiers far beyond the range of metric maps.
personalrobotics.github.io/lrn/
🧵1/6
🤖 Real-world tested: LRN cuts down interventions on Spot and a large tracked vehicle.
âś… Plug & play: Works with nearly any local stack that accepts goal waypoints.
🔄 Auto-labeled: Trained from raw FPV videos using CoTracker to trace camera paths.
🧵4/6
This work is a collaboration between the Personal Robotics Lab (@siddhss5.bsky.social) and Robot Learning Lab at the University of Washington @uwrobotics.bsky.social @uwcse.bsky.social
🧵5/6
This project was a fun effort with Matt Schmittle, Nathan Hatch, Rosario Scalise, @mateoguaman.bsky.social, Sidharth Talia, @khimya.bsky.social, @siddhss5.bsky.social and Byron Boots.
🧵6/6
🔥 Key insight: Robots can reason further by learning to identify distant affordable frontiers as intermediate goals.
🧠How it works: It uses a pre-trained SAM2 backbone + small head to find frontiers in images. Given a goal, it selects the highest-scoring one to navigate to.
🧵3/6
Natasha Jaques
If you visited the @uwcherryblossom.bsky.social, did you “spot” an unusual visitor among the blooms? Researchers in the @uofwa.bsky.social #UWAllen #robotics group recently took advantage of some nice weather to take our Boston Dynamics robot dog for a stroll around campus. #AI 1/5
UW Robotics
Rohan Baijal
Rohan Baijal
Rohan Baijal
Rohan Baijal
Rohan Baijal
Rohan Baijal
Rohan Baijal
University of Washington researchers created a method for training AI systems — both for large language models like ChatGPT and for robots — that can better reflect users’ diverse values. It...
