Pier Abbeel's Berkeley Robot Learning Lab releases WARP-RM, using timescale warping to filter noisy demonstration datasets

Original post

Data quality is too often the unsung hero for deep learning. Excited to finally share WARP-RM, where we train a model to learn the rate of task progress, rather than absolute progress, by warping the timescale of existing trajectory data. Check out @uynitsuj 's thread for more! 👇

Justin Yu ✈️ ICML 🇰🇷@uynitsuj

Introducing WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation 🧵

We gave our t-shirt folding robot more demonstrations and it got worse. Every extra demo ended in a successfully folded shirt. The data wasn't bad. It was noisy. The policy couldn't tell productive motion from dead time, and it imitated both equally. So which moments of a demo are actually worth copying?

🌐 Project Website: https://uynitsuj.github.io/warp-rm 📄 Paper: https://arxiv.org/abs/2606.28320 💻 Code: https://github.com/uynitsuj/WARP-RM 📨 XDOF blog post: https://xdof.ai/blog/warp-rm

10:22 AM · Jun 30, 2026 · 2K Views

warp rm

UYNITSUJ.GITHUB.IOVia

WARP RM

GITHUB.COMVia

Philipp Wu@philippswu

Train a WARP-RM yourself using our ABC dataset!

11h4323

LIKES4REPLIES1

Philipp Wu@philippswu

Better data results in better models. WARP-RM extends XDOF's capabilities in data understanding to help us provide the highest quality data to our customers!

You can checkout our blog post here as well! https://www.xdof.ai/blog/warp-rm

11h2854

RETWEETS1

Ken Goldberg@Ken_Goldberg

Excited about WARP-RM: by modulating the time-sampling on a few high-quality human demonstrations, we train a progress measure to extract high-quality segments from all demos, yielding significant improvement in folding throughput. Hats off to Justin and the team for this idea!

Justin Yu ✈️ ICML 🇰🇷@uynitsuj

Introducing WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation 🧵

🌐 Project Website: https://uynitsuj.github.io/warp-rm 📄 Paper: https://arxiv.org/abs/2606.28320 💻 Code: https://github.com/uynitsuj/WARP-RM 📨 XDOF blog post: https://xdof.ai/blog/warp-rm

1h3810