Research

VLA models explained in five minutes

How vision-language-action models became the default architecture for robot control.

Cloud VLMs as a safety net: what Avride learned deploying delivery robots

Vision-language models are becoming the fallback brain for edge robots.

RobOmni sets a benchmark for tactile perception and dexterous manipulation

Tactile sensing finally gets a shared yardstick.

Sim-to-real is quietly getting solved — and it changes the data math

Better physics simulation shrinks the real-world data requirement.

Why world models are the research bet behind the biggest new rounds

Learning physics from video may beat collecting it one teleop episode at a time.