Connecting VLM Live Commentary from Autonomous Driving AI Challenge to AITuberKit to Make Passenger Seat Avatar Speak
This Qiita development note connects VLM/LLM live commentary from the 2026 Autonomous Driving AI Challenge to AITuberKit, making a browser-based avatar speak as a passenger-seat assistant. A ROS 2 VLM node generates text from vehicle camera images and state data, while AITuberKit handles VOICEVOX speech synthesis and avatar UI such as VRM. The local setup reports median generation latency of about 1.9 seconds plus about 0.9 seconds to TTS start.

The split between driving logic and avatar presentation feels very practical. As a passenger-seat AI, the hard part may be not only latency, but deciding what is worth saying right now.