Upstream Expands Live Digital Twins From Connected Vehicles to Robots and Physical AI

A warehouse robot running hotter than usual could be developing a mechanical fault—or simply working harder in a warmer building. A drone deviating from its normal pattern could be responding to its mission, a software problem, or a cyberattack. As autonomous machines spread into more industries, knowing what a machine is doing becomes inseparable from understanding why.

Upstream is bringing its experience monitoring connected vehicles to that problem. The company has announced an expansion into the broader physical AI market, extending its live digital twin infrastructure to robots, humanoids, drones, autonomous mobile robots and agricultural equipment.

The announcement builds on infrastructure that Upstream says has been proven across more than 40 million connected assets. That figure describes its existing connected-asset footprint, rather than a newly deployed fleet of robots. The company’s proposition is that lessons learned from automotive security and operations can help manufacturers manage autonomous machines as they move into commercial deployment.

From Connected Cars to Autonomous Machines

Connected vehicles provide a useful starting point because they combine software, sensors, remote services and real-world consequences. A problem can originate in a component, an application, a communications interface or the relationship between them. Diagnosing it requires more than looking at an isolated error message.

Robotics brings similar complications. A machine’s behavior depends on its environment, workload, configuration and software. An alert that makes sense for one fleet may be misleading for another. As deployments grow, operators need a way to maintain that context across individual machines and groups of assets.

“Connected and autonomous vehicles are among the most advanced Physical AI systems operating at commercial scale today,” said Yoav Levy, Upstream’s co-founder and CEO, in the announcement. He said the company is extending technology and expertise developed with automakers to robots, humanoids, drones and other autonomous machines.

Upstream says it will continue expanding its automotive business alongside the new physical AI practice. The move broadens the environments its platform addresses, bringing its existing monitoring approach into additional categories of connected equipment.

What a Live Digital Twin Actually Adds

In this context, a digital twin is a continuously updated representation of a machine’s behavior and operating state, together with its surrounding digital ecosystem. It does not have to be a photorealistic 3D replica. Its value comes from connecting signals that would otherwise sit in separate systems.

For the expansion, Upstream describes combining telemetry, operational state, mission context, software activity and anomalies. Those inputs can help an investigation distinguish behavior associated with an intended task from changes that warrant attention.

The company’s platform documentation adds detail about the underlying data pipeline. Incoming information is normalized and cleansed, with a common dictionary helping reconcile different data sources. The platform includes privacy capabilities such as anonymization and tokenization, and combines rules for known threats with machine learning that identifies unfamiliar patterns.

That architecture matters because fleets rarely generate perfectly uniform data. Different hardware generations and software versions can describe the same event differently. Before an AI system can investigate a change, it needs a consistent account of which asset produced the signal and what else was happening around it.

Upstream says the infrastructure is hardware- and protocol-agnostic and does not require software agents installed on the devices themselves. It can be deployed in a customer’s cloud environment. The practical dependency remains access to useful telemetry: an agentless architecture still needs data sources and integrations that reveal the behavior operators want to understand.

Where Machine Learning and AI Agents Fit

Upstream’s cybersecurity platform describes Ocean AI as combining machine learning for anomaly detection, generative AI for investigation and agentic AI for response. These are distinct jobs. Detection identifies unusual activity; investigation assembles and interprets relevant evidence; response connects findings to an operational workflow.

For physical AI, that sequence could help teams investigate an unexpected change across a fleet before deciding whether it calls for maintenance, a security response or further observation. Those are illustrative uses of the architecture, rather than disclosed results from a named robotics customer.

The distinction also clarifies what Upstream is announcing. This is infrastructure for monitoring, understanding and managing connected machines. The release does not introduce a robot foundation model, a new manipulation policy or a system that independently teaches a robot to perform a physical task.

Four Uses for the Same Operational Context

Upstream identifies cybersecurity, quality and safety, operations, and compliance as the principal applications of its expanded platform. Each depends on understanding machine behavior, but each asks a different question of the data.

Security teams need to recognize potentially malicious activity and investigate its impact. Engineering teams need to identify degradation, software anomalies and emerging faults. Operations teams need visibility into fleet health, utilization and uptime. Compliance teams need evidence they can inspect and use in reporting or audits.

A shared behavioral record can help connect those efforts. For example, an engineering anomaly and an unusual remote interaction might deserve examination together, even if separate teams initially receive them. Whether that connection improves a particular deployment will depend on data coverage, detection quality and the way findings enter existing workflows.

The Regulatory Timing Requires Precision

The expansion also arrives as the European Union’s Cyber Resilience Act begins imposing reporting obligations. According to the European Commission’s reporting guidance, those obligations took effect on September 11, 2026. Manufacturers of products with digital elements that fall within the rules must report actively exploited vulnerabilities and severe incidents affecting product security. An early warning is required within 24 hours of becoming aware, followed by notification within 72 hours.

The Commission’s CRA summary distinguishes that reporting timetable from the regulation’s broader application on December 11, 2027. Scope and applicable exclusions matter; a device’s classification as a robot or a physical AI system alone does not establish every obligation.

Monitoring can support incident detection, investigation and evidence gathering. It cannot by itself establish compliance. Manufacturers still need processes for assessing incidents, making notifications and meeting the requirements that apply to their products.

What to Watch as the Platform Expands

The announcement does not provide pricing or identify named customers for the new robotics applications. Those details, along with evidence from real deployments, will help show how well an automotive-derived approach transfers to machines with different operating environments and data constraints.

The central opportunity is clear: autonomous machines create a growing need for systems that can explain their behavior after deployment. Upstream is betting that the digital infrastructure used to understand connected vehicles can become part of that operational foundation. Its next test is whether the same contextual intelligence can give robotics teams useful answers quickly enough to improve security, reliability and day-to-day fleet performance.