bf16 breaks VLA fine-tuning: one of the month's most interesting robotics papers
m_wulfmeier · x · 2026-11-18
Markus Wulfmeier highlights a paper showing that bf16 precision can break VLA (vision-language-action) fine-tuning, calling it one of the more interesting papers this month — a notable finding on how numeric precision affects robot foundation model training.
More from Embodied
- Singapore's Agnes AI demos one AI coordinating drones, hoppers and robots in a Mars rescue — Scobleizer · 2026-11-18
- Oru'el claims Kanook beats NVIDIA Flora and OpenAI on RoboRewardBench by 30-50% — AashaySachdeva · 2026-11-18
- Voxel Robot Hails a Taxi, and Roboticists Are Excited — chris_j_paxton · 2026-11-18
- HT-Policy: multimodality isn't why diffusion policies win at imitation learning — chris_j_paxton · 2026-11-18
- Team builds on-device robot dog in 3 hours with phone-side Gemma, fully offline — DynamicWebPaige · 2026-11-18
- UMA Robotics bets on Europe first: survey shows EU edges US on talent, customers, cost — RemiCadene · 2026-11-18