Bodhi — the humanoid that answers

UBTech Yanshee YOLOv8 NeMo ASR Wake-word routing

Ask it, and it answers out loud

You say Bodhi's wake word, ask a question, and it looks at what's in front of it and answers by voice. The clip has sound — the reply you hear is the robot's own, generated from what it recognized in the moment.

It runs on a UBTech Yanshee, a small hobby-grade humanoid. Turn the sound on.

How it works

Vision runs on YOLOv8, so it can name the objects in view instead of guessing.
Speech comes in through NVIDIA NeMo recognition; a wake word gates it, so the robot only acts when it's actually being addressed and ignores the rest of the room.
All of it runs on a UBTech Yanshee — a small hobby-grade humanoid. The hardware is modest on purpose; the work is the software that ties seeing, listening, and answering into one loop.

What broke, honestly

Reserved for the failure log: the wake-word false triggers, the latency fights, and how they got tamed. Write-up in progress.