Robot papers dominate the window. The strongest pattern is practical control: Vision-language-action (VLA) models get motion pretraining, runtime correction, persistent world models, and explicit injury tests.
Frozen VLA deployments now have concrete add-ons to test: execution-time action selection for chunked policies, refusal tests for hazardous robot videos, and geometric memory for world-model rollouts used in policy…
The day’s robotics papers focus on making Vision-Language-Action (VLA) policies execute reliably under real deployment conditions.
VLA teams can make three concrete changes to current robot policy work: add geometry-conditioned action decoding for fine manipulation, run a local latent-prompt adaptation pass before rollout, and store replay memory…