Servoservobrief.com, Wednesday, August 12, 2026
A model-and-deployment day: DeepMind put whole-body control in a robot foundation model while Dexterity and Agility pushed real jobs forward.
-
Google DeepMind releases Gemini Robotics 2 with whole-body control
DeepMind's new robot foundation model (an AI that maps camera input and instructions to robot motion) now coordinates the whole body, legs and arms together, not just a fixed arm. A companion model, Gemini Robotics ER 2, adds video understanding and multi-robot coordination. No numbers disclosed and no independent benchmark yet, so treat capability claims as demo-stage.
Read more on Google DeepMind -
FedEx expands autonomous trailer loading with Dexterity at a Maryland hub
Dexterity widened its multi-year FedEx deal to a second site, the Hagerstown hub, for robots that load trailers on their own. This is a production deployment, not a pilot, and one of the few humanoid-adjacent programs with a repeat customer scaling to more facilities. No unit count or contract value disclosed.
Read more on Dexterity -
Agility opens a Fremont facility and appoints a CFO ahead of its SPAC merger
Update: since Agility's June deal to go public via Churchill Capital, it has opened a new Fremont, California development site and named Michael Beer CFO, the standard staffing for a public company. A CFO plus a US facility near Tesla signals the SPAC is moving toward completion. No new financial terms disclosed.
Read more on Agility Robotics -
London's Humanoid becomes Europe's first humanoid-robot unicorn with a $152M round
UK startup Humanoid raised $152M, pushing its valuation to about $1.35B, the first European humanoid maker to cross the billion mark. It shows the capital wave now extends past the US and China poles, though the company has shown no independently verified deployments yet. Figure vendor-reported, early stage.
Read more on Funding wire -
Enigma raises $71M for simpler robot teleoperation control
Enigma raised $71M to build controls that make steering a robot as simple as a volume dial, aimed at the teleoperation layer that feeds robot training data. Easier human control means cheaper data collection, the bottleneck most learning-robot startups are chasing. No deployment figures disclosed.
Read more on TechCrunch