# Building the systems behind the next generation of intelligent robots > Source: > Published: 2026-08-26 09:37:15+00:00 # Building the systems behind the next generation of intelligent robots Robotics is entering a new phase. Advances in AI are expanding what robots can perceive, reason about, and interact with in the physical world. This is helping to move robotics beyond demonstrations and into real-world deployment across manufacturing, logistics, healthcare, agriculture and more. [According to the International Federation of Robotics](https://ifr.org/ifr-press-releases/news/top-5-global-robotics-trends-2026), the global market value of industrial robot installations has reached a record USD $16.7 billion, and this momentum is expected to accelerate as robots move into more industries and take on more complex real-world tasks. For robotics companies, the challenge is now scaling robots from impressive demonstrations to commercial systems that can operate safely, reliably, and economically in the real world. Every new AI capability adds pressure on compute, software, power efficiency, cost, safety and system integration. This raises a fundamental systems challenge: **How can we make intelligence reliable in machines that must sense, move, adapt, and operate safely in real time?** Federico Pecora, Senior Principal Robotics Research Lead at Arm, explores this in a new technical paper called * The missing science of robotic systems*. The paper sets out a system-wide approach for the next phase of [physical AI](https://newsroom.arm.com/blog/what-is-physical-ai), showing why real-world robotics depends on bringing AI inference, sensing, planning, control, safety, and orchestration together inside machines operating in the real world. ## What robots require to operate in the real-world A robot is a continuously operating autonomous system. It combines camera feeds, lidar, radar, force sensors, motor control, AI inference, planning, safety monitoring, and real-time feedback as it moves through changing physical environments. These workloads operate at different rates and may need to run concurrently, each posing different compute requirements. Motor control needs predictable timing, while perception and AI inference need high-throughput compute. Planning needs access to the robot’s sensor data and understanding of its environment while safety monitoring needs priority and isolation. At the same time, all these functions need to share data without creating delays, bottlenecks, or unpredictable behaviour. Keeping these workloads running together seamlessly is one of the biggest engineering challenges in modern robotics. As robots take on more complex tasks, engineering teams need to understand how each capability affects the full system. For example, a larger model, richer world model, or more continuous reasoning layer may improve task performance, but also increases demands on memory, bandwidth, scheduling, and runtime control. ## The foundation beneath robots and physical AI All robots – whether that’s industrial, autonomous vehicles, humanoids or service – need systems that can place, schedule, isolate, and coordinate different workloads while keeping real-time control and safety-critical functions predictable. [Each category places different demands on compute](http://newsroom.arm.com/blog/the-evolution-of-physical-ai-from-controlled-environments-to-the-real-world), yet all require a system that can coordinate intelligence and action under real-world constraints. This makes workload placement a central design decision. Engineering teams need to decide what runs on the CPU, what is accelerated, what belongs in a real-time domain, what data needs to move between functions, and which workloads need priority or isolation. While accelerators execute AI models, CPUs commonly coordinate much of the surrounding system. They manage sensors, schedule workloads, move data between processors, maintain the operating system and ensure safety-critical software continues running even when AI workloads become more demanding. Meeting the demands of next-generation robotics requires a compute foundation that enables performance, efficiency, safety and software portability across the full spectrum of workloads. Arm aims to help reduce complexity and accelerate innovation in robotics, building a deeper understanding of robotic systems and architectures to help engineering teams bring new capabilities from research concepts towards deployable products. ## Building the future of robotics on Arm The next phase of robotics will be shaped by systems that can turn AI capability into reliable physical performance. Robots will need to sense, reason, and adapt to complex environments while maintaining predictable control, efficient power use, and safe operation. Arm is helping make that future possible by giving the robotics ecosystem the compute foundation built for heterogeneous, power-efficient, and scalable systems. From high-performance application processing to low-power sensing, control, and safety-critical execution, Arm enables engineering teams to build robots that can operate with greater intelligence, efficiency, and reliability in the real world. Federico Pecora’s latest technical paper examines the emerging science of capability integration, and the questions compute architects must address as physical AI moves into deployed systems. The opportunity ahead is significant: moving robotics from impressive capability to dependable deployment, enabling engineers to build machines that can sense, move, adapt, and act safely in the real world at scale. #### Building physical AI for real-world robotics Explore Arm’s new technical paper of how AI inference, sensing, planning, control, safety, and orchestration interact and what those interactions may imply for the compute architecture of future robots. Any re-use permitted for informational and non-commercial or personal use only.