ETH Introduces Vernata: Label-Free Self-Supervised Learning for Outdoor LiDAR
rsasaki0109 · x · 2026-08-11
Researchers from ETH Zürich introduced Vernata, a multi-modal, multi-teacher distillation framework for self-supervised learning on outdoor LiDAR point clouds without requiring labels.
- Core Extensions: Building upon the Sonata architecture, it introduces sparse view augmentation for robustness against varying point densities, a memory bank mechanism to stabilize training, and cross-modal distillation using high-resolution 2D image features for fine-grained semantic guidance.
- Results: Significant performance improvements over baselines, achieving mIoU scores of 54.7 on TartanGround (+12.1%) and 57.1 on Waymo (+14.7%).
- Reduced-Modality: The approach maintains strong performance even in reduced-modality settings (lacking color or normals).
More from Embodied
- SpaceX Starship and Tesla Optimus Could Be the Key to Building Lunar Bases — XFreeze · 2026-08-11
- Asimov 1 Humanoid Robot to Begin First Deliveries This August — freelerobot · 2026-08-11
- Alibaba's RynnValue robot value model uses temporal distance to boost real-world success to 72.5% — Alibaba-DAMO-Academy · 2026-08-11
- Show HN: I Vibecoded a Heavy Lift Drone — liseman · 2026-08-11
- The Robot Taking Your Job Won't Be Impressive, Just Cheaper — VraserX · 2026-08-11
- 200ms Cloud Latency is Fatal for Robots: Why Edge Inference is Mandatory — vaibhavbetter · 2026-08-11