Research [ 04 ] Learning in the loop

Continual Learning Through Embodied Interaction

Embodied interaction is a source of experience for studying how AI can learn from people's responses and feedback.

Published: 2026-08-21
Author: Chen Li

FIG. 01The loop, as designed

The question: How can interaction improve continually through people’s responses and feedback?
Where it stands: The loop is designed and partly built. No public artefact yet shows one real interaction completing it end to end — from the interaction, through a recorded decision, to a changed policy running on the robot.
What the figure is: A designed diagram of the loop we are building. It is not a recording of one, and no step in it is a measurement.

Learning is one part of our research into embodied social intelligence. We use robots as a concrete platform to explore how experience from interaction could improve subsequent understanding and behaviour.

Communication brings understanding, collaboration produces experience, feedback corrects behaviour. All of it feeds one loop: Interaction → Experience → Learning → Capability.

We chose to lay the foundation first, rather than use a set of canned reactions to fake a robot that already understands people.