93 “The real challenge isn’t building one smart robot. It’s operationalising physical intelligence at scale” INDUSTRIALS & MANUFACTURING typically required significant reprogramming or engineering effort. What is changing now is the emergence of AI systems capable of generalising across tasks, supported by advances in simulation and cloud-scale infrastructure. Chien points to a convergence of three forces: foundation models, high-fidelity simulation environments and scalable cloud platforms, which, together, enable a new kind of industrial intelligence. For manufacturers, this means robotics is no longer just an automation tool, but part of a broader AI system capable of continuous learning and adaptation. “The shift is that robots are becoming adaptive systems that can perceive, reason and act in real-world conditions that aren’t fully predictable,” says Chien. “What’s driving it is the convergence of foundation models that generalise across tasks, simulation environments mature enough to validate physical behaviours before deployment, and cloud platforms that orchestrate the full lifecycle at enterprise scale.” While intelligent robotics demonstrations are increasingly common, Chien says the industry’s focus is shifting from individual machines to large-scale deployment. The challenge is ensuring physical intelligence can operate reliably across fleets of robots, multiple facilities and diverse vendor ecosystems while maintaining safety, governance and ongoing optimisation. “That’s an enterprise platform challenge, not just a robotics challenge,” he says. A major bottleneck is data. Unlike digital environments, real-world robotics data is expensive, fragmented and difficult to capture safely at scale.
RkJQdWJsaXNoZXIy NzQ1NTk=