Representative End User Clients
Representative Automation Clients
Representative Software Clients
In our first post, we introduced the concept of Physical Intelligence—the fusion of AI and robotics that marks a new era for industry. The concept is exciting and transformative, but as an analyst who has covered the industrial edge so closely, my first question to Colin was this: where does the thinking happen? An AI model running in the cloud is a powerful tool, but for a robot navigating a dynamic environment, latency isn't just a performance metric; it's a safety and operational necessity. When an AMR is moving at two meters per second, the time it takes for a signal to travel to the cloud and back is an eternity.
This becomes even more critical in the context of human-robot collaboration. For a cobot working inches away from a person, a two-second delay in recognizing an unexpected human movement is an unacceptable safety risk. It’s a liability, not a solution. For robotics, and especially for cobots, the answer must be at the industrial edge. The "thinking" has to happen on the robot, in real time.
This is the physical foundation—the bedrock—of Physical Intelligence. It’s not just about bolting a computer onto a robot; it's about designing a holistic, ruggedized nervous system capable of real-time perception, reasoning, and action. In my ongoing research into the hardware that enables this new generation of intelligent robots, I'm seeing the emergence of a new class of compute platforms purpose-built for this task. We're talking about industrial-grade, GPU-accelerated systems-on-modules (SoMs) and embedded PCs, such as NVIDIA’s Jetson line, that are designed to run complex AI models in harsh, vibration-prone environments with limited power—right on the machine.
This explosion in edge compute capability is enabling a vast diversity of robotic forms. While the media focuses on humanoid robots, the real innovation will be in applying this intelligence to the most effective form for a specific industrial task, whether it has wheels, legs, or is a stationary arm. It’s about matching the compute to the optimal physical function, not just a human-like form.
This local, high-performance compute is what makes real-time perception and action possible. A constant, massive stream of data from LiDAR, 3D cameras, and force/torque sensors—the "eyes and ears" of the robot—can be processed right on the device. This is a critical discipline known as sensor fusion, where multiple data streams are integrated to create a single, high-fidelity model of the robot's immediate surroundings. This allows the robot to make the split-second decisions required for safe and effective operation.
Furthermore, as these robots become more intelligent and connected, they also become a significant new attack surface. My work on OT cybersecurity highlights that securing these autonomous endpoints is paramount. These aren't just IT devices; they are physically acting agents integrated with production systems. A compromised cobot is not just a data breach risk—it’s a potential safety and operational catastrophe. As industry incorporates new technology and workflows for robotics, we must apply the same cybersecurity rigor and standards, like IEC 62443, to a fleet of mobile robots as we do to a plant's distributed control system.
Colin’s Perspective: It's fascinating, Patrick. From my software-centric view, what you're describing is the physical manifestation of the perception layer. Without the performant, secure edge compute hardware layer, the AI models I focus on would be starved of the real-time data they need to function. The best AI in the world is useless if it's blind, slow, and insecure. You're building the hardware stack that allows the "State"—the real-time situational awareness—to even exist within an AI's context window.
Exactly. Before we can talk about the "ghost," we have to build and secure the "machine." In our next post, Colin will take this physical foundation and explore how his work on data fabrics and AI provides that ghost—the intelligence that turns a machine into a truly collaborative, context-aware agent.