From HumanUse to Crowd Mesh

The Mountain at 2:00 AM

An ultramarathon on a remote mountain ridge at two in the morning exposes the real bottleneck of physical work. When storms hit, the people scrambling over wet rocks to keep runners safe are local volunteers. Their limitation is never willingness; it is situational context. When an agent in a lightweight lens provides quiet spatial cues, ad-hoc human effort transforms into a coordinated mesh. Agency is not a matter of replacing human muscle, but of removing human blindness.

Following our hackathon proof of concept with Rokid and Antigravity, HumanUse showed what happens when the AI acts as the brain and the human acts as the actuator. The question is how to scale this closed loop from one person to a collective.

Five Grounded Realities

Beyond the controlled setting of a demo corridor, physical friction defines five immediate arenas:

1. Disaster search grids. In collapsed ruins where cellular networks fail, head-mounted sensors log cleared sectors across a local mesh. Voluntarily swept terrain updates in real time, lifting the physical fog of war without duplicate searches.
2. Ten-second guided triage. When facing sudden injuries, panic is the enemy. On-device reasoning walks a volunteer through basic AVPU checks—Alert, Voice, Pain, Unresponsive—stabilizing victims before medical teams arrive.
3. Dignified elder care. In caregiving, robots feel cold and distant. Placing the lens on the caregiver keeps the human touch and eye contact intact, while the agent quietly flags medication schedules, subtle mood changes, or mobility hazards.
4. In-situ handy repairs. Fixing a complex pump or wiring harness should not require scrubbing greasy fingers across a phone screen to find a tutorial video. The agent identifies the exact valve or wire in view and shows the next physical move.
5. Ad-hoc event marshaling. Directing hundreds of runners through mountain fog requires shared awareness. Course marshals receive localized route and safety cues directly in their line of sight, turning dispersed volunteers into an adaptive safety crew.

The Real-to-Sim Inversion

Sim-to-real tries to force gaming logic into physical machines. Real-to-sim does the opposite: it uses wearable sensors on human actuators to turn real-world chaos into a cooperative game. The hardest part of physical agency was never physical power or motor speed—it was situated perception. Locomotion and hand skills took millions of years of evolution to develop; the human body is already an exceptional actuator.

The Cooperative Game

This collapses the artificial boundary between physical AI and gaming AI. They are two sides of the same coin. One trains algorithms inside game engines; the other uses agents in lightweight optics to lift the physical fog of war for human teams. By connecting a central planner with human hands and feet, we do not automate people away. We elevate everyday volunteers into a synchronized crowd mesh, turning messy physical crises into solvable spaces.