Simplex Robo Attn: helping robots choose where to look

A robot has detected a person. Where should it look next?

In real interaction, relevant subjects change continuously: someone speaks, enters the scene, gestures for attention or turns away. Simplex Robo Attn, our active robotic visual-attention effort, brings the question of where to look into the interaction loop.

From a fixed view to active attention

A fixed camera provides an observation window, but an interaction partner may move away from its center. Active attention needs to balance continuity with awareness of new events.

The goal is not to turn toward every change. Excessive switching can interrupt a conversation. We explore maintaining attention on meaningful evidence, observing again when uncertain and releasing attention when an interaction ends.

Working with P0

P0 supplies signals about people, gaze, speaking activity and actions. Robo Attn explores how these signals can guide observation. Together, they can help the robot both interpret what it sees and seek information needed for its next judgment.

Illustrative scenarios include considering a person who appears ready to engage, maintaining attention on the current partner in a group, and reassessing attention after that partner leaves. These describe our intended interaction scenarios, not universal success-rate guarantees.

Attention communicates

Orientation can express that a robot is listening. Unnecessary tracking or abrupt turns can instead feel disruptive. We therefore consider perception quality, switching stability and the effects of physical movement together.

We will share demonstrations of target selection, sustained attention and multi-person interaction. This introduction leaves internal control parameters and implementation details for later.

Demo in preparation: establishing attention, following a moving partner and switching after an interaction ends.