KDDI and ELYZA set up ELYZA Robotics for physical AI

ELYZA Robotics brings together experts from across the KDDI group who had been working on physical AI for over a year, consolidating them into a core organization. The entity aims to combine ELYZA's foundation model research with KDDI's infrastructure and operational know-how. ELYZA has also set up ELYZA RSI Research to study automated research and AI-driven AI development, and the two leaders see potential to apply that work to physical AI.

A major focus is data collection. The executives noted that deciding which sensors to use, such as RGB versus depth, and how much full-body data to capture, requires iterative improvement. They consider field-specific datasets a more durable technical asset than individual model architectures. On the hardware side, robots in research use break down within a few months, and the lack of a repair ecosystem means operations can be disrupted. Benchmark results also show a gap: the CVPR 2026 winning model achieved about 41% success, with second place and below at about 21% or less, and the current Table30 V2 leader is in the low 40% range.

The organization plans to start deployments in settings like the Lawson store inside KDDI's office and eventually expand to logistics, manufacturing, and hospitals. KDDI's data center in Sakai is expected to support low-latency connections to large models, which the executives see as a way to combine real-time responsiveness with larger model sizes.