At Tokyo Big Sight today, KDDI Corporation and AVITA Inc. are demonstrating a humanoid robot at KDDI's booth (P8, South Exhibition Hall) at Google Cloud Next Tokyo, the search giant's flagship Japan cloud conference running July 30 through 31. The robot is the first in the country to run on Gemini 3.1 Flash Live Preview, Google's end-to-end audio model built to preserve emotional nuance in real time, according to KDDI's official announcement. Unlike previous approaches, the model is native audio-to-audio, ingesting raw speech and producing audio output without converting voice to text and back, recognizing "acoustic nuances like pitch and pace" in real time. This contrasts with SoftBank's Pepper, launched in 2014 and halted from production in 2020 after approximately 27,000 units were manufactured, which could not reliably recognize emotional cues. Where the previous API-bridged approach produced response delays of "several seconds," the new architecture effectively eliminates that lag, collapsing the "wait-time stack." AVITA, founded in June 2021 by roboticist Hiroshi Ishiguro, a Distinguished Professor at Osaka University's Intelligent Robotics Laboratory, did not arrive at physical humanoids directly.