The most underpriced line in every autonomy budget is data, and Toyota wants to manufacture its way around it. Toyota Research Institute's September 21, 2021 grant US11126891B2 claims simulating sensor data using a generative model — synthetic training data instead of fleet-collected miles.
Read the economics in the claim. The CPC tags — G06N 3/0454 generative networks, G06K 9/6257 and G06K 9/6289 training-data handling, G06F 30/20 simulation, G06T 19/003 synthetic scene rendering — describe a data-factory, not a sensor. The asset being built is cheaper training input.
The decoder problem is that real-world data collection is enormous and largely invisible in financials. Fleets of instrumented vehicles, armies of labelers, petabytes of storage — it is buried in aggregate R&D. Synthetic data is a direct attack on that buried cost, which is why it is strategically valuable even if it never shows up as a revenue line.
For a fundamentals reader, the lever is cost structure. An autonomy program that can substitute generated data for collected data bends its development-cost curve down. The patent is Toyota staking a claim on that lever in 2021.
The honest limit: a synthetic-data patent does not prove the data is good enough to replace real miles, nor does it quantify Toyota's savings. Sim-to-real gaps are stubborn. The grant signals intent to compress the data bill, not proof that it worked.
The takeaway for the money desk: the autonomy cost story is increasingly a data-cost story. Watch which players are patenting synthetic-data pipelines — they are the ones trying to fix the most expensive, least-discussed line in the budget.