Growing up, I watched Star Wars and was fascinated by C-3PO and R2-D2 in those science fiction movies – the idea of lifelike machines, so-called humanoid robots interacting with humans, sparked my imagination. Little did I know those childhood dreams would become a reality sooner than expected. “I wouldn’t be surprised if we are the last generation for which those sci-fi scenes are not a reality,” says Alexander Khazatsky, a machine-learning and robotics researcher at Stanford University in California, mentioned in the article ‘’The AI revolution is coming to robots: how will it change them?’’. Today, humanoid, autonomous robots are no longer confined to our imagination. They are real, and they're revolutionizing how we live and work with the help of advanced computing and artificial intelligence. 

The humanoid robot is the most complicated type of robot within the robotic categories. Building the brain, or compute, of a humanoid means running perception, reasoning and hard real-time motion control on the same platform, without any of them stealing time from the others. For physical AI, consistency under data load matters as much as raw throughput, and that changes the compute decision from “Which chip has the most TOPS?” toward “Which architecture lets perception, reasoning and deterministic control run together, every cycle reliably?”

A humanoid runs two minds at once

A humanoid system architecture is structured into three main layers: high-level AI for reasoning, mid-level motion control for deterministic balance and kinematics, and low-level real-time actuator feedback. Each layer runs at its own rhythm and puts a different demand on the silicon it is running on. Current practice has been to distribute the three layers on two different embedded systems:

• System 1 (deliberative): High-level AI for reasoning – understanding the scene and planning the next move.

• System 2 (reactive): Mid-level motion control for balance and low-level embedded control for real-time actuator feedback

This separation has traditionally been required because the different layers work with fundamentally different data structures and compute algorithms. The high-level reasoning is characterized by throughput-oriented batched math, whereas the reactive layers need strict latency and have to process branch-heavy real-time tasks. However, this is not necessarily the case anymore with heterogeneous computing architectures. 

Mapping the workload to the engine

GPUs are excellent at the high-level AI zone for the massively parallel perception and vision-language models. But a robot must plan, coordinate and control in real time, and that work belongs ideally to a CPU. With the rise of heterogeneous computing architectures, designers don’t need separate systems anymore, as described above. Today, all that is required is one single processor: The AMD Ryzen™ AI Embedded X100 Series offers all the required compute engines in one package. It features up to 16 high-performance “Zen 5” CPU cores for real-time tasks, the discrete-class AMD Radeon RDNA 3.5 GPU, and a low-latency AI engine XDNA2 NPU for highly parallelized workloads with shared unified memory. In practice, this means:

• The AI runs on the GPU and NPU.
• Motion control and real-time feedback run on uniform “Zen 5” CPU cores that behave identically cycle after cycle so control loops stay deterministic and on time.

Modern humanoids may simultaneously run multiple camera streams, LiDAR, speech, local LLMs, VLMs, manipulation models, and navigation maps. This makes data acquisition and management challenging, and when the controlling CPU cores do not provide sufficient performance, scheduling and synchronizing can cause unwanted latency. And here is where the AMD Ryzen AI Embedded X100 series excels. It features up to 16 high performance Cores with up to 5.1 GHz frequency, reducing latency to a minimum. Plus: with shared unified memory, the CPU, GPU, and NPU can access the same data directly, reducing latency compared with discrete CPU-GPU systems that must transfer data via PCIe architectures.

ROCm and the open software stack: the other side of the coin

Hardware is only one side of the coin. For a robotics engineer, the silicon is worth exactly as much as the toolchain and libraries that exposes it — and this is where AMD’s open source software approach plays in. ROCm is AMD’s open-software stack for GPU-accelerated computing. It gives robotic designers a software stack – including drivers, libraries, compilers, and AI frameworks – that allows robot developers to use AMD GPUs for computationally intensive tasks such as computer vision, SLAM, motion planning, simulation, and machine learning. ROCm enables GPU acceleration for frameworks like PyTorch and TensorFlow, which can significantly improve inference speed and parallel processing performance in autonomous systems.

From silicon to system: the module

As robots become smarter, more autonomous, and face more requirements for interacting with the physical world, the winning designs will be the ones whose processors work together to meet the real-time demands of the "sense-decide-act" loop. congatec offers heterogeneous platforms like the AMD Ryzen™ AI Embedded X100 Series on a standards-based Computer-on-Module. This module gives robotic developers a strong foundation in a single validated and standardized building block. This is a decisive factor as processor technology and AI acceleration evolve and OEMs want to scale and implement the next generation in their designs. With a simple module exchange they can easily upgrade to the next performance level, without a complete re-design.

The conga-HPC/cRX1 is built on the COM-HPC form factor, designed to deliver the performance, bandwidth, and high-speed I/O interfaces required by increasingly demanding robotic workloads. By combining heterogeneous compute with a modular, scalable platform, robotic developers can spend less time solving the underlying compute challenge and more time building robots that can sense, decide, and act in the real world, just like C-3PO and R2-D2!

May the force be with you!