Won 3rd place at the MakeMIT x Harvard hackathon this weekend after ~24 hours of building hardware at MIT.
Built a small robotic swarm called Triad.
The idea was: give AI agents physical bodies (and describe capabilities as tools), make a few of them and see what happens when they coordinate.
Each robot ran an OpenClaw agent on a Raspberry Pi 5.
Movement, vision, and speech were exposed as tools the agent could call. They could see, follow people, talk, and decide when to act.
Feels like there’s a real gap around agent-to-agent communication and how physical capabilities are represented.
That’s the part I want to keep digging into.