Physical AI is redefining modern robotics, drones, and industrial systems, where machines must perceive, decide, and act in real-time. In this paradigm, intelligence is not measured by model size or TOPS, but by latency, energy efficiency, and instant response. Automaton Engine is developing an energy-efficient silicon chip that enables fast, reliable decision-making under strict power constraints, enabling machines to act in real time.
For such applications, AI systems must make decisions in less than 50 milliseconds, often directly at the point of use and without offloading to the cloud. At the same time, only 5 to 30 watts are usually available at the edge. Conventional approaches often use excessive numerical precision, such as FP16 or FP32, and consequently consume more energy than necessary.
At the JARA Institute for Energy-efficient Information Technology (PGI-10) at Forschungszentrum Jülich, researchers are therefore working on the Automaton Engine, an edge AI system for reliable real-time decision-making. The aim is to develop a real-time-capable and energy-efficient AI chip for the robotics industry that strengthens the competitiveness of AI chip design in Germany, supports a sovereign European AI stack and enables energy-efficient automation on a large scale.
The approach is based on a patented low-bit AI engine with full-stack control. It utilises proprietary low-bit precision processing with AFP-5, achieving near-FP32 model accuracy, and can be used via an end-to-end, PyTorch-compatible toolchain from training through to deployment. The key factor here is not maximum computing power in TOPS, but optimisation for latency per watt.
The solution includes software, firmware and hardware, as well as an FPGA demonstrator and an overview of hardware silicon development. Automaton Engine thus addresses a growing market for physical AI systems and targets applications where fast, reliable and energy-efficient decisions are required directly at the point of use.
Find out more at: www.automatonengine.com
Group Leader
