Yilun Du: Energy-Based Models Enable Robot Generalization by Construction
du_yilun · x · 2026-09-09
Yilun Du shares his research perspective 'Generalization by Construction' on moving robot generalization beyond collecting more demonstrations.
- Robots can use inference to solve tasks they were never explicitly trained on — planning actions and goals before acting with learned world models
- For placing a mug on a crowded table, learning to evaluate whether a candidate placement is stable and reachable relies on simpler, more reusable relationships than the full scene-to-placement mapping
- Inference compares and refines candidate placements; energy-based models (EBMs) provide a natural landscape for this search
Related event: Du Yilun Proposes Reasoning-Driven Robot Generalization(3 posts)→
More from Embodied
- π0.5 study: raising demos from 10 to 438 lifts Place Ball in Bowl success from 7% to 83% — stingning · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- HandsOn benchmark: 66 verified tasks and 18,000 human demos for dexterous manipulation — stingning · 2026-10-27
- DexPort: a handheld interface with a real robot hand for dexterous-hand data collection — pabbeel · 2026-10-27
- ROS names Alejandro Hernández Cordero new project lead, eyes 2027 release — ___Mufasaa · 2026-10-27
- ROS Meetup coming to Google's NYC office Oct 29 with speakers from Tangent, Cohesive Robotics, Intrinsic — kscottz · 2026-10-27