DeepMind Robotics Accelerator Opens With Gemini ER-1.6 Access

On June 9, 2026, Google DeepMind launched the Google DeepMind Accelerator: Robotics, a three-month initiative based in London targeting early-stage European startups. The program bridges the gap between frontier research and market-ready Physical AI by opening access to the company’s proprietary embodied AI infrastructure.

Infrastructure and Program Mechanics

The 12-to-15-week equity-free program uses a hybrid format, starting with five days of in-person technical workshops before moving to remote training and mentorship. To qualify, teams needed verified venture backing and at least five members with deep in-house AI expertise, enforcing a technical baseline across the cohort.

The accelerator’s primary value proposition is hardware and model access. Participating startups receive up to $350,000 in Google Cloud credits and direct integration support for Google’s Gemini robotics models. This includes the base Gemini Robotics model and the extended reasoning variant, Gemini Robotics ER-1.6. This stack allows early-stage companies to bypass the capital-intensive phase of pre-training foundational vision-language-action models and focus directly on task-specific fine-tuning and hardware integration.

The 2026 Cohort

The initial cohort focuses heavily on industrial, healthcare, and logistics applications. The accepted companies demonstrate a shift away from bespoke control systems toward generalized foundational models applied to specific hardware profiles.

Notable participants include:

• 3D-Components AS: The Norwegian startup is developing RobTrack, an automated platform for robotic welding and metal 3D-printing. The system accelerates parameter selection by a claimed 280x compared to traditional methods.
• Acumino: A Greek firm building hardware-agnostic Physical AI control systems designed to improve ROI for existing industrial hardware fleets.
• Adapta Robotics: Based in Romania, the team is deploying physical AI equipped with human-like touch sensors for automated quality assurance in the automotive and healthcare sectors.

DeepMind’s Hardware Strategy

The accelerator follows a rapid sequence of industrial AI deployments for Google DeepMind in the first half of 2026. Rather than building commercial hardware, the company is positioning its Gemini model suite as the default operating system for third-party robotics.

In January 2026, DeepMind partnered with Boston Dynamics to deploy Gemini models into the Atlas humanoid robot. Two months later, the company signed a strategic agreement with Munich-based Agile Robots SE to scale foundation models across industrial platforms. DeepMind is also currently constructing a fully automated laboratory in the UK, utilizing Gemini-powered robotics to synthesize hundreds of novel materials daily.

For hardware developers, the accelerator highlights a structural shift in robotics engineering. As model providers commoditize the underlying control and reasoning layers, developers building physical AI agents will find competitive advantage moving toward proprietary datasets, specialized sensor arrays, and highly constrained domain workflows.