VLS: training-free VLM steering of frozen robot policies wins CoRL and CVPR workshop awards
micoolcho · x · 2026-09-29
RoboPapers teased a podcast on VLS (Vision-Language Steering), a CoRL 2026 paper from UW, Oxford, NUS and Allen AI that also won Best Paper Runner-Up and Outstanding Paper awards at CVPR 2026 workshops. VLS is a training-free framework that adapts frozen diffusion/flow-matching robot policies at inference time: a VLM generates differentiable rewards for partially denoised action proposals, steering sampling so policies handle out-of-distribution objects, scenes, and instructions without fine-tuning. Code and arXiv are public.
More from Embodied
- ProHand Gen 2 robot hand launches at IROS 2026 with 2M-cycle durability and +66% abduction — chris_j_paxton · 2026-09-29
- Figure CEO Brett Adcock says goodbye to the F.02 humanoid robot — adcock_brett · 2026-09-29
- Self-supervised multisensory pretraining robot RL paper wins best student paper at IROS — GeorgiaChal · 2026-09-29
- Meta Connect hackathon: apps built in hours that once took teams weeks — HarperSCarroll · 2026-09-29
- Earth Rover Challenge: robots run policies in the wild in Costa Rica, dodging curious kids — micoolcho · 2026-09-29
- VkVIO: First cross-platform GPU-accelerated VIO using vendor-agnostic Vulkan — zhenjun_zhao · 2026-09-29