The VLA Feedback Loop Closes: From Touch to Policy in One Step

Research

Tactile VLA: Closing the Sense-Think-Act Loop for Manipulation

This paper tackles a huge gap: most VLAs are blind to touch. It's a direct move towards robots that can feel and adapt, not just see and plan.

Fast VLA: Learning to Act Faster with Efficient Action Prediction

Speed is deployment. This work on making VLAs run faster is pure leverage for builders trying to get from simulation to a real, responsive robot.

Learning Robotic Policies from Imperfect, Noisy Demonstrations

Finally, a paper that doesn't assume perfect data. This is tactical gold for anyone working with real-world, messy human demos.

Tools

NVIDIA Isaac Sim: New Tools for Synthetic Data & Domain Randomization

NVIDIA keeps sharpening the sim-to-real pipe. The new tools for generating better synthetic data are a direct path to reducing real-world data collection costs.

Events

ROSCon 2025 Agenda Drops: Keynotes on Scaling and Real-World Deployment

The schedule is out, and the focus is clear: scaling and shipping. This is the conference for builders, not just researchers.