Hyundai Motor unveils on-device AI chip for robots

Hyundai Motor Group’s Robotics LAB has announced a new on-device AI chip for robots, which will be produced by South Korea-based semiconductor firm DEEPX under a three-year partnership.

The chip, developed with control software from Robotics LAB and hardware from DEEPX, enables robots to process information and make decisions autonomously without cloud connectivity.

Hyundai Motor Group said the chip will help address labor shortages and industrial automation challenges.

This “edge brain” chip uses less than 5 watts of power, and is designed for real-time recognition and control in both industrial and everyday settings.

The technology is being tested in Hyundai’s DAL-e Delivery robot and a facial recognition system called Facey, with plans for broader use in security and mobile robots.

The partnership aims to support future mass production of robots, and expand applications in sectors like logistics, hospitals, and airports.

🔗 Source: Hyundai Motor

🧠 Food for thought

Implications, context, and why it matters.

Chip specs matter beyond the partnership news

DEEPX’s DX-M1 (a neural processing unit for edge AI) delivers 200 GPU-class TOPS at 2-5W, and DX-M1M provides 25 TOPS in the same range 1 2. Hyundai’s “less than 5 watts” claim does not say which tier it plans to use or if this is a new variant.
Missing details on memory bandwidth and process node leave a gap. Native support for Robot Operating System 2 (ROS 2), an open-source robotics middleware standard, is unclear. That blocks comparisons with NVIDIA Jetson Orin Nano (a widely used edge AI module for robots) or Qualcomm’s RB5 platform (a reference platform for robotics).
DXNN supports Open Neural Network Exchange (ONNX) and TensorFlow model deployment 3. Hyundai has not said whether its software plugs into standard robotics frameworks or depends on proprietary tools, which could deter third-party developers.

An open ecosystem would let system integrators build robotics deployment services

Releasing full developer kits and ROS 2 support could mirror NVIDIA’s accelerated ROS 2 packages for Jetson platforms 4. That would let integrators offer optimization and deployment for logistics and healthcare operators.
The Hyundai deal runs three years, so mass production likely comes later. That window could let independent software vendors (ISVs) build safety certifications because DEEPX chips ship with a Root of Trust (a cryptographic foundation that verifies device integrity) and a True Random Number Generator (TRNG) 2.
Vendors building ROS 2 toolchain integration could target sites that run delivery robots or security systems. That hinges on DEEPX confirming SDK support for standard simulation environments and publishing detailed compiler docs beyond the current ONNX conversion pathway 5.