Bardienus Duisterhof

@bardienus

Incoming Postdoc @stanford || PhD Student @CMU_Robotics with @jeff_ichnowski || Efficient World Models for Robot Manipulation 📷 ||

USA

bart-ai.com

Replies Reposts Media

Pinned
Bardienus Duisterhof
@bardienus
Sep 17
Introducing PointZero—a 3D world model pre-trained without robots. Dexterous manipulation requires understanding diverse 3D dynamics, but current approaches rely on expensive robot data. So how can we pre-train a 3D dynamics model without it? PointZero introduces a simple idea:
00:00
8
Bardienus Duisterhof
@bardienus
Sep 23
Today I defended my PhD! 🎓 I came to CMU five years ago to pursue my career, but I’m leaving most grateful for everything I’ve learned about myself along the way. Thank you to everyone who has been part of this journey, including, of course, my committee: @jeff_ichnowski
33
Bardienus Duisterhof
@bardienus
Sep 17
Check out @Arthur_K_J 's work on efficient dynamics models for precise, dynamic manipulation! Turns out a specialist model does the job sometimes :)
Arthur Jakobsson
@Arthur_K_J
Sep 17
Humans learn by interacting with the world: How can we teach robots to similarly adapt to ambiguous dynamics? Introducing Wiggle and Go! #CoRL2026. We find that a specialized model can learn rope properties from just one interaction, and complete complex dynamic tasks. The robot
00:00
Bardienus Duisterhof
@bardienus
Sep 16
World models typically reason about the future with video models, but is video really the best modality? Each task may need different modalities to perform best; what if we predict all of them? Really excited to release ModAR today: autoregressively predicting the future as a
Adam Hung
@Adamjhung
Sep 16
World-action models typically imagine the future in RGB – but are pixels really the right representation for robotics? Our bet is no: RGB spends capacity on fine-grained details and variation that are often irrelevant for robot policies. DINO features, point tracks, and depth
00:00
2
Bardienus Duisterhof
@bardienus
Sep 8
Atlas learns to predict RGB-D frames. Can we directly predict 3D tokens instead? Check out our work 'SPAR3S', a sparse voxel-aligned 3D latent generative model for scene completion. SPAR3S does not require ground-truth 3D data for supervision; instead, it supervises a sparse
00:00