● Humanoid Fleet Control Center

A simulated car assembly line run by 12 humanoid robots. The control center assigns every task, decides when robots charge or get serviced, and reacts to disruptions you trigger. Switch strategies to see how much the orchestration matters. All figures come from this simulation, not a real plant.

Strategy benchmark: one simulated shift-hour

Runs each strategy for 60 simulated minutes on 3 random seeds, with the same scripted disruptions in each run: an engine shortage at minute 10, a 30-second protective stop at Powertrain at minute 25, and two robot faults at minute 40. The current fault-rate setting applies.

Turning this into a real Claude Code control center

Fleet layer → MCP serverEach robot or cell reports its state and accepts commands through an MCP server. In a real plant this would wrap the fleet manager, ROS 2, VDA 5050 for mobile robots, or OPC UA for the line PLCs.
Fast loop stays deterministicTask bidding and charging rules (the scheduler in this page) run in milliseconds in ordinary code. Claude sets the policy but doesn't sit inside the real-time loop.
Claude as supervisorClaude handles the exceptions: re-sequencing around a parts shortage, re-planning after a fault, slotting in a rush order. It explains each decision to the shift lead. Subagents can each own a cell.
Safety is hardwareProtective stops come from safety PLCs, light curtains and robot controllers. The AI is told about a stop and plans around it, but it never starts or overrides one.

Orchestration strategy

Inject a disruption

Line metrics

Control center console

Robot inspector

Click a robot on the floor to inspect it.