I Gave an AI a Body — Cyrus Clarke, MIT Media Lab
An MIT prototype shows embodied agents need an interaction language, low-latency feedback, and human validation—not just a model connected to actuators.
Cyrus Clarke connected an OpenClaw agent to a **900-pin shape display** and let it create programs for the device. A later closed-loop system generated and evaluated movements until it retained **32 usable gestures** for body-language responses.
For physical or ambient agents, design a small validated gesture vocabulary that can respond before slower language generation finishes. Keep machine scoring in the iteration loop, but require people to confirm that expressions are legible and appropriate.
Cyrus Clarke connected an OpenClaw agent to a **900-pin shape display** and let it create programs for the device. A later closed-loop system generated and evaluated movements until it retained **32 usable gestures** for body-language responses. For physical or ambient agents, design a small validated gesture vocabulary that can respond before slower language generation finishes. Keep machine scoring in the iteration loop, but require people to confirm that expressions are legible and appropriate. This remains an exploratory Media Lab prototype, not evidence of consciousness or general robotics capability. The gesture system was unpublished at the time of the talk, and the transcript gives no formal user study, latency measurement, or safety evaluation.
This extends agent interfaces from screens and speech into physical expression, while narrowing the practical design problem to a small, validated gesture vocabulary. The closed scoring loop supports iterative discovery, but the candidates reinforce that machine evaluation cannot replace human checks for legibility, timing, appropriateness, and safe actuation.