Intelligence Must Flow from Sensing and Perception to Inference, Planning, Control and Ultimately Physical Action

In this conversation with Vaishali Umredkar, Editor, Semiconductor For You, Rohith Gopalakrishna, Country Sales Manager, Embedded Group, AMD India, explains how AMD’s CPUs, GPUs, NPUs and adaptive computing are enabling intelligent, autonomous robotics. He highlights heterogeneous compute, real-time processing and physical AI, describing the architecture as a “distributed nervous system” for next-generation intelligent machines.

How is AMD’s portfolio of CPUs, GPUs and adaptive SoCs addressing the growing computing needs of robotics?

The robotics industry is entering a new phase as systems evolve from traditional automation toward intelligent, autonomous machines that can perceive their surroundings, make decisions and act in real time. This is creating increasingly diverse computing requirements from sensing and perception through decision-making and control. These systems must deliver real-time responsiveness, deterministic behavior, functional safety and reliability, while operating within strict power and thermal constraints and remaining resilient when connectivity is limited or unavailable.

At AMD, we bring together CPUs, GPUs, NPUs, FPGAs and adaptive computing across our portfolio, so developers can apply the right type of compute for the right workload. That heterogeneous approach helps balance performance, responsiveness and power efficiency as robotic systems become more capable and compact.

ADVERTISEMENT

One way to think about this architecture is as a distributed nervous system: CPUs and GPUs provide the “brain” for complex reasoning and perception; Adaptive SoCs and NPUs function like the nervous system, enabling low-latency, power-efficient processing close to where decisions must be made. FPGAs support precise control of sensors and actuators, allowing intelligent machines to safely and accurately interact with the physical world.

How are AMD Ryzen AI embedded processors and Versal adaptive SoCs enabling AI and real-time processing in robotics?

Our x86 embedded processor portfolio is designed to address different compute needs across an intelligent robotic system. AMD Ryzen AI Embedded X100 and P100 Series processors combine CPU, GPU and NPU capabilities to support AI inference, computer vision and real-time processing at the edge. The X100 Series, for example, is designed for demanding autonomous systems where deterministic real-time control and low-latency AI processing are critical. AMD EPYC Embedded processors provide scalable, server-class compute and I/O for demanding industrial edge workloads. Versal adaptive SoCs can be deployed close to sensors and actuators to support functions such as sensor fusion, data processing and deterministic control. Together, these platforms give developers flexibility to distribute workloads across the system based on what the application requires.

What roles do AMD GPUs and AI accelerators play in enabling advanced vision, perception and autonomous decision-making in robots?

GPUs and dedicated AI accelerators like the NPUs are central to enabling the vision, perception and AI inference workloads required for increasingly autonomous robots. AMD GPUs can handle demanding computer vision workloads, including multi-camera processing, while NPUs can efficiently run inference workloads such as object detection and scene understanding. Our approach also enables CPU, GPU and NPUs to work together across the workload, helping minimize unnecessary movement of data and support low-latency processing. This is important because physical AI extends beyond running AI models. Intelligence must flow from sensing and perception through inference, planning, control and ultimately physical action. By combining different compute engines across this pipeline, developers can match the right compute to each stage while helping reduce latency and improve responsiveness as robots interpret their surroundings and respond in real time.

How is AMD helping robotics developers balance AI performance, power efficiency and real-time processing at the edge?

For robotics developers, balancing performance, power efficiency and responsiveness becomes more important as more AI workloads move to the edge. AMD addresses this through heterogeneous compute and workload consolidation, bringing multiple processing functions together in integrated platforms. This can help reduce the number of discrete components required and improve power and thermal efficiency. Our heterogeneous approach that combines CPU, GPU, NPU and adaptive computing supports different workloads – from AI perception and inference to real-time control.

A key part of this is the AMD Robotics Software Suite, which provides an open software foundation for developing and deploying physical AI applications across AMD platforms. Built around widely adopted frameworks and tools including PyTorch and ONNX, the suite helps developers bring together AI perception, reasoning, orchestration and real-time robotics workloads while maintaining flexibility across the hardware stack. This gives developers a more integrated path from development and prototyping to deployment, while the AMD Robotics Partner Network extends the ecosystem with hardware, software, simulation, perception, and system integration partners.

The goal is to give developers the flexibility to optimize their systems for the performance, power, and real-time requirements for their application.

How does AMD see India contributing to the development and adoption of AMD-powered robotics and intelligent automation?

India is an important market for the development and adoption of intelligent automation, with a strong technology ecosystem and growing demand for AI across industries. We are seeing opportunities across manufacturing, healthcare, automotive and industrial automation, where AI and edge computing can address increasingly complex workloads. As India’s AI ecosystem continues to grow, we see an opportunity for the country to contribute to both the development and deployment of intelligent automation globally.