Boston Dynamics is injecting Google DeepMind intelligence into its Spot robot platform for improved autonomous machinery inspections.
To give the hardware autonomous reasoning capabilities tailored specifically for heavy industry, Boston Dynamics partnered with Google Cloud and Google DeepMind to integrate Gemini and Gemini Robotics ER 1.6 into Orbit AIVI-Learning, delivering a more sophisticated, intuitive, and powerful AI experience.
Because of this integration, Boston Dynamics Spot and Orbit now continuously learn about facilities with unprecedented depth, allowing for higher-order reasoning and more complex visual analysis. Spot can now visually read analogue gauges, detect hazardous chemical spills, and interpret the physical environment without constant human oversight.
AIVI-Learning delivers insights across multiple stakeholders to give operators a holistic view of what is happening on site. This allows facilities to accumulate value across multiple operational areas:
- Safety and security: AIVI-Learning executes EHS checks such as looking for dangerous debris or spills. This helps reduce fines, risk, and potential liability.
- Asset monitoring: The AI performs inspections of key assets like conveyor belt damage, sight glass levels, and gauges to prevent critical failures. Monitoring these key assets prevents downtime.
- Materials and 5S: The system identifies material movement throughout a facility and automates manual inspections previously conducted by multiple people across many shifts. It effortlessly handles advanced tasks like 5S compliance audits, accurate pallet counting, measuring sight glass fullness from 0-100%, and detecting puddles of standing liquid.
IEEE Spectrum reports the scale of this deployment is already massive. Several thousand of these hardware platforms currently patrol industrial sites globally.
Spot evaluates its surroundings, recognises its own computational limits, and automatically calls on external AI tools when it encounters an anomaly it cannot independently process. Furthermore, the system gets smarter behind the scenes with Zero-Downtime Upgrades. These sophisticated AI models are continuously updated and refined in the cloud, meaning inspection accuracy improves automatically without requiring the team to run manual Orbit software updates or schedule downtime.
The financial weight behind physical AI
A recent analysis from Treble PR noted that funding for physical AI – encompassing robotics, computer vision, and agentic intelligence – has reached $26.7 billion. Investors are pouring cash into companies bridging the gap between digital models and physical space.
Large language models process text brilliantly, but heavy industry requires physical intervention. Mining conglomerates, automotive manufacturers, and oil refineries need systems that can navigate complex three-dimensional spaces.
Boston Dynamics is already proving how AI models can expand these robotic capabilities. In a 2025 hackathon, developers experimented with Google’s visual-language model (VLM) Gemini Robotics-ER 1.5 to empower Spot with embodied reasoning. Rather than writing formal software logic, developers interacted with Gemini Robotics using conversational language.
Using Spot’s SDK, they developed a layer that facilitated interaction between Gemini Robotics and Spot’s application programming interface (API). Gemini Robotics was given a finite set of tools to control the robot, which translated inputs from the AI into actual API calls. This allowed the AI to evaluate images from Spot’s cameras, identify target objects, and dynamically sequence actions.
When Spot finds a spill, it cross-references the visual data against chemical hazard databases, alerts the floor manager, and initiates containment protocols through connected industrial software APIs.
To maintain safety, Gemini Robotics operates with strict boundaries. The AI cannot invent new capabilities or control Spot beyond what is available through the API, keeping the robot’s behavior predictable while still allowing it to adapt to different situations.
Tiered intelligence also provides a safeguard. The DeepMind integration allows the robot to flag uncertainty. If environmental conditions degrade – say, steam obscures a pressure gauge – Spot stops, documents the obstruction, and pings a human operator. The system knows what it does not know.
Enabling next-gen predictive maintenance
Fixing a machine before it breaks can save millions in unplanned downtime. A single Spot unit replaces hundreds of static sensors and walks predetermined routes; scanning thousands of components per shift. It uses thermal imaging to detect overheating transformers and acoustic sensors to hear the hiss of a compressed air leak.
Data hygiene also improves instantly. Human inspectors can have inconsistent reporting, while Spot takes the exact same photo, from the exact same angle, under the exact same lighting conditions, every single day. This creates an impeccable historical dataset.
Machine learning models thrive on structured, consistent data. By feeding Spot’s daily visual audits into an overarching AI platform, plant managers can track microscopic degradation over time. They can watch a crack form in a pipeline days before it ruptures.
Labour dynamics play a massive role in this adoption curve. Heavy industry faces an aggressive talent shortage. Veteran technicians are retiring, taking decades of intuitive, undocumented knowledge with them. Younger workers show little interest in manual factory floor inspections so Spot executes the dull, dirty, and dangerous tasks.
The engineer’s role is already shifting toward setting goals and objectives, while the multi-modal robot foundation model interprets those instructions to form complex and adaptive plans for Spot to execute. The pressing question is whether legacy enterprise networks can handle the immense data load required to keep these autonomous systems alive.
See also: Gartner: Half of new warehouses will be ‘human-optional’ by 2030
Want to learn more about the IoT from industry leaders? Check out IoT Tech Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including AI & Big Data Expo and the Cyber Security Expo. Click here for more information.
IoT News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.