Google's embodied-reasoning model for robotics
Google on July 30, 2026 announced Gemini Robotics ER 2, an "embodied reasoning" model designed to act as a high-level brain for robots, coordinating perception, planning, and tool use while delegating motor execution to lower-level vision-language-action (VLA) models. The company described the system as enabling robots to chat with humans, understand the physical world, and plan multi-step tasks, with the ability to natively call tools such as Google Search and other user-defined functions. Google positioned ER 2 as a significant upgrade over its predecessor, Gemini Robotics ER 1.6, with continuous video understanding that lets robots track their own progress, self-correct when something goes wrong, and decide when to move on to the next step.
Dexterity, multi-robot collaboration, and developer availability
Axios reported that the broader Gemini Robotics 2 release bundles models that control individual physical actions with those handling higher-level task planning, quoting Google's blog post: "Gemini Robotics 2 enables robots to reason through every movement, unlocking a broad range of tasks." Google cited examples including humanoids walking, crouching, stretching, and manipulating objects to clean a cluttered room, as well as multiple robots teaming up to finish jobs faster. Multi-robot collaboration is introduced as a new capability, allowing robots to work together in shared spaces and complete complex workflows a single robot could not do alone. Gemini Robotics ER 2 is publicly available to developers via the Gemini API and Google AI Studio, with a private preview on the Gemini Enterprise Agent Platform.
Latency-sensitive orchestration and a Boston Dynamics demo
To address the speed demands of physical-world reasoning, Gemini Robotics ER 2 integrates into the Gemini Live API through a bidirectional streaming endpoint optimized for latency-sensitive tasks, which Google said allows fluid orchestration in which the model commands action models and robotics APIs to complete multi-step tasks without "stop-and-think" pauses. To illustrate the design, Google built a demo with Boston Dynamics' Spot quadruped, using Gemini Robotics ER 2 to orchestrate Spot APIs including navigation and manipulator movement so the robot could fetch objects on request.
BYD readies first humanoid robots for August debut
Separately, China's largest electric-vehicle maker BYD told the South China Morning Post that its first humanoid robots would debut in August at the company's Di Space experience centres, where, according to a statement issued on Thursday, the robots would be more than a concept product and capable of interacting with visitors. The Shenzhen-based automaker, which has repeatedly signalled ambitions in embodied AI, is positioning the launch as entry into a humanoid arena where Tesla has already fielded its Optimus robot. In June, BYD executive vice-president Stella Li said the company had made "a huge investment in AI" and related technologies such as humanoid robotics, and predicted that China would become the first market to "see full commercialisation of humanoid robots."
Regulatory backdrop and competitive landscape
Axios framed Google's software push against a tightening hardware environment, noting that US companies may face greater difficulty obtaining the robots needed to run such systems after the FCC earlier in the week moved to ban future sales of Chinese-made robots on security grounds. Together, the Google and BYD announcements illustrate how the sector is being shaped simultaneously by US AI-model advances and by Chinese automakers extending their electric-vehicle expertise into physical AI, with Tesla's Optimus serving as the most-cited reference point in the new wave of humanoid launches. The next verifiable milestone is BYD's planned August debut at its Di Space centres, alongside broader public rollout of Gemini Robotics ER 2 through Google's developer channels and further implementation steps for the FCC's Chinese-robot import restrictions.
Share this article
