Better AI models and local reasoning alone will not make intelligent machines dependable. They also need connectivity to sense beyond their own hardware capabilities and collaborate with others. That connectivity must be secure and available when needed, even when network conditions deteriorate or adversaries interfere.
Physical AI closes the loop between sensing, reasoning and action. A robot, autonomous vehicle, drone or industrial machine observes its environment, interprets what it sees and then changes the physical world. That final step makes technology fundamentally different from an AI service that only produces information.
Autonomy under pressure: mines, drone swarms, robots and defence
Consider the vision of an autonomous underground mine. A loader must navigate tunnels, exchange positions and intent with other vehicles and remain visible to the control room. Private 5G/6G and edge computing can support mobility and local coordination, but the machine must still stop safely if every external link fails. If an attacker can spoof its position or alter a route, connectivity becomes a hazard rather than an advantage.
Now think about an autonomous drone swarm mapping a wildfire or a rescue search mission in a disrupted area. Mesh or device-to-device communication can support local coordination, 5G/6G can connect the swarm to edge services, and satellite can provide backhaul beyond terrestrial coverage. Yet one rogue node or false observation can corrupt the common operating picture. Secure joining and removal, mutual authentication, encrypted group communication, detecting fake base stations, resilient positioning and behaviour-based detection are essential functions.
The reliable connectivity challenges are yet more difficult in defense use cases, where systems must be prepared for adversarial interference such as jamming and spoofing, and use special connectivity technologies with a low probability of being detected, intercepted or observed. Technologies for this purpose include AI-optimised frequency hopping schemes, very narrow “pencil beam” radio connectivity, reconfigurable intelligent surfaces (RIS) and free-space optical connectivity.
Consider a care robot assisting nurses and patients in a hospital. To move through crowded corridors safely or help someone out of bed, it must continuously interpret movement, voice, position and other sensor data. These decisions cannot wait for a distant cloud service: multisensory AI processing and safety-critical inference must take place locally, close to the robot. Secure, low-latency connectivity complements the robot’s own observations with information from staff, medical systems and other equipment. The robot must continue to operate safely even if the connection is lost. In care settings, dependability means, above all, that people can trust the machine to work safely beside them.
On a factory floor, in a warehouse or at a construction site, the focus shifts to collaboration between people and robots. To cooperate safely, the robot must continuously interpret human movement, gestures, voice and its surroundings while anticipating what its co-workers may do next. Connections with other machines, workers and workplace systems support this collaboration. The aim is to enable people to work naturally and safely with robots, not merely around them.
What should Physical AI product developers do now?
These example missions point to a shared design priority: developers should define communications and trust architecture at the same time as AI and safety architecture. Set explicit latency, availability and recovery requirements, and define how the system will recover from disruptions. Use multiple communication links, or bearers, where the operational risk justifies them. Apply threat modelling, hardware roots of trust, secure software supply-chain practices and signed updates from the first prototype. Plan the migration to quantum-resilient cryptography from day one and ensure that cryptographic algorithms can be updated over time.
Most importantly, validate the combined cyber-physical system under realistic stress. Test congestion, handovers, jamming, spoofing, credential theft, malicious updates and loss of cloud or positioning services. A system tested only in ideal network conditions is not ready for the physical world.
Dependable autonomy requires local intelligence and resilient connectivity
The architecture behind these recommendations must balance the benefits of connectivity with the need for local autonomy. One often heard misconception deserves to be challenged: connectivity does not mean sending everything to the cloud. A machine cannot assume that a certain network will always be available, that latency will remain low or that a remote service will respond in time. Safety-critical decisions must be made locally, at the edge of the network. Connectivity should extend the machine’s awareness and capabilities, not become a single point of failure.
The right architecture combines local intelligence with several complementary communication links. 5G-Advanced and upcoming 6G can provide managed mobility, positioning and predictable service quality. Wi-Fi and optical connectivity support high-capacity or deterministic communication locally. Direct 5G side link, Mesh and device-to-device connections allow machines to coordinate when infrastructure is unavailable. Satellite and other non-terrestrial networks extend coverage to remote areas.
The design principle is simple: if the primary data link disappears, the system should hand over to another bearer or enter a safe local mode. It should never move from intelligent operation to uncontrolled behaviour.
When software controls motion, cybersecurity becomes safety
Alongside resilient connectivity, the control loop must also be trustworthy. A spoofed position, manipulated sensor reading, malicious software update or hijacked command channel can produce a physical incident. AI-based systems introduce new vulnerabilities across the AI value chain. This is why cybersecurity must be built into every stage of the product lifecycle, rather than added at the gateway just before deployment.
Every machine, sensor, operator and software workload needs a verifiable identity and least-privilege access. Devices need secure boot, signed firmware and protected over-the-air updates. Communications require encryption, careful key management and crypto-agility, including a migration path towards post-quantum cryptography. Runtime monitoring must detect sensor manipulation, connectivity spoofing, denial of service and compromised AI components.
Europe should compete on dependable, high-trust systems
Europe has deep capabilities in microelectronics, chips, communications, edge computing, cybersecurity and high-performance computing. The opportunity is to combine them into industrial systems that are safe, scalable and resilient in real operating environments. This is also where VTT can make a distinctive contribution: bringing advanced connectivity and cybersecurity technologies, partners and test environments together so that promising components become validated solutions.
The winning physical AI products will not simply have the most impressive standalone AI model. They will remain coordinated, trustworthy and safe when the environment changes, networks deteriorate and attackers interfere. That is the difference between autonomy and dependable autonomy, and it is the standard we should aim at.
Physical AI and future connectivity are rapidly converging: intelligence is moving closer to machines, networks are becoming distributed AI platforms, and development is shifting from individual technologies towards complete, continuously validated systems. VTT can help turning this transition into sustainable growth of companies by combining its expertise in secure connectivity, edge AI, sensing, cybersecurity, microelectronics and photonics with real-world pilots and neutral testing, simulation and inference environments. By helping companies move faster from promising technologies to dependable, commercially scalable solutions, VTT can renew established industries, enable new products and businesses, and strengthen Europe’s competitiveness, resilience and technological sovereignty.
Ari Aalto is Vice President, Digital Infrastructures at VTT. He works at the intersection of secure connectivity, cybersecurity and emerging digital systems, helping translate research capabilities into industrial and societal impact.
Petteri Annamaa is Connectivity Technology Portfolio Lead at VTT. He works with advanced connectivity hardware and software systems, creating new innovations for connected machines, people and organizations.