bdambrosio look-at-target
Closed-loop gaze (Tier 2). Given a natural-language target, the tool drives the
ChatterBot head until the target is centered in the camera, then returns the
final pose and the centered frame (which you can then display to the user).
How it works (all internal — one step from your view):
- Capture the current view and judge whether the target is visible.
- If not visible, sweep a small set of safe poses, capturing at each, until the target appears (or give up after a bounded search).
- Once visible, repeatedly judge the target's position relative to frame center and nudge pan/tilt toward it, bounded per step and clamped to the safe head envelope, until centered or an iteration cap is reached.
It is deliberate and slow (each cycle is a move, a settle, a capture, and a visual judgement) and it centers once — it is not a continuous tracker. If the target keeps moving it returns its best effort. The returned frame is attached to your view so you can confirm what you ended up looking at.
Examples
{"thought": "the user asked me to point at them", "tool": "look-at-target", "target": "me, the person talking to you"}
{"thought": "find and face the cat", "tool": "look-at-target", "target": "a cat"}