Camera reliability scoring for VLAs: frozen model, no heavy fine-tuning
paraschopra · x · 2026-10-27
Resident Aryaman studies how VLA robot models break under camera degradation: in early experiments with two global cameras and a wrist camera, noise on even one camera significantly reduced task success. His fix: a lightweight camera-quality estimator that scores each camera's reliability and uses the scores as attention biases so the model favors better views. The pretrained VLA stays frozen; only the estimator trains on task loss with a small dataset. He also has a parallel AI4Science project designing lipid nanoparticles for RNA drug delivery to solid tumors.
More from Embodied
- Ben Lorica changes his mind on world models: if AI builds the robot, where's the value? — bigdata · 2026-10-27
- Toyota Pours $6.4B a Year Into Humanoid Robots, Targets 400,000 by 2028 — emmanuelvivier · 2026-10-27
- GPT-6 moves a camera to insert a flower, yet burns an egg: reflections on the Astra report — liuziwei7 · 2026-10-27
- Harness Device ditches paper box for premium UV-printed metal tin packaging — dee_hw · 2026-10-27
- $299 Desktop AI Assistant Intern 2 Adds iMessage, Ships Preloaded With OpenClaw Agent — dee_hw · 2026-10-27
- UBTECH opens humanoid robot factory producing one robot every 10 minutes — import_jmr · 2026-10-27