Each point of the coating carries a damage value D(x,y) ∈ [0,1]: the fraction of the ~100 µm coating stack (40 µm clear coat, 30 µm base coat, 30 µm primer) that is gone (0 = intact, 1 = down to bare steel). One grid cell is 12.5 mm wide. A robot sits idle until the damage it senses exceeds θ, then it steers toward the worst cell inside its 120 mm sensing radius and heals everything inside its 42 mm repair radius:
activate ⇔ D(x_robot) > θ (or worst D in sense radius > θ)
dD/dt = −0.9 · k · n_active(x, t) for D > 0
steering: x' = x + (x_target − x)(1 − e^(−3 dt))
n_active is the number of active robots whose repair disc covers the cell. The ideal heal time of one robot on a peak of depth A is t = (A − 0.01)/(0.9 k); the sim uses a fixed 1/60 s step, so the checks accept one step (16.7 ms) of discretisation error. This page runs the identical physics module as the 3D version, so both give the same numbers.
- What the 2D view adds: the cross-section strip shows which layers are lost along the dashed line (clear coat first, then base coat, primer, steel), and the graph shows health and active robots over the last 60 s of simulated time.
- Real technology it echoes: embedded sensor networks that monitor body panels for micro-damage and self-healing coatings whose microcapsules (amber dots in the strip) break on a scratch and release a polymerising agent. Here mobile robots replace fixed capsules.
- Scale is exaggerated: robots are drawn about 10⁷ times larger than a real nanoscale device, and the cross-section is stretched vertically (not to scale).
- Repair energy is illustrative: healed volume × an assumed 12 mJ/mm³, a budget proxy for harvested power, not a measurement.
- Simplifications: first-order steering lag (no actuator physics or collisions), linear healing in time, a scalar damage field without crack mechanics.
- Not affiliated with any car maker; the door is a generic test skin. Art rendered in Blender (Cycles).