Robot learning work centered on making policies executable under real control constraints. ZR-0, T2VLA, and Chronos show the main emphasis: cross-embodiment supervision, reward-free test-time improvement, and memory…
Robot policy teams can get more value from the current work by adding execution-facing checks around evaluation, deployment, and geometry.
This period’s robot papers focus on making vision-language-action (VLA) policies usable after deployment. InSight adds new manipulation primitives through robot rollouts. Reflective VLA records action consequences.
Robot VLA teams can now test three practical support layers around existing policies: a primitive acquisition loop for missing manipulation skills, a geometry path for multi-camera fine-tuning with limited real demos…
April 9 centers on embodied models that make structure explicit. The clearest pattern is morphology, future state, and object kinematics being written into the learning problem itself.
The clearest near-term build is a training pipeline that uses generated futures to write navigation labels offline, then deploys a small student model alone.