Models in the loop

AI & Embodied Intelligence

Foundation models, VLAs, and control policies moving into hardware.

3 entries

A vision-language-action model that runs on-device

The team distilled a 7B parameter policy down to something that fits in 8GB. Trade-offs Success rates drop roughly four points against the full model.

SIG//73657d agoPHYSICAL INTELLIGENCE

Physical Intelligence releases a cross-embodiment dataset

The release covers twelve robot morphologies and about 900 hours of teleoperated manipulation.

SIG//25907d agoGOOGLE DEEPMIND

DeepMind reports sim-to-real transfer without fine-tuning

Policies trained entirely in simulation held up on hardware in 61 percent of trials.