The TEMPEST paper does not deeply discuss robustness in adversarial scenarios (such as deliberately mimicking another person's driving style), and the scale of 45 drivers remains limited relative to actual large fleets.