What do you do with interventions even in massive fleet operations? Rebalancing training data by harvesting correct fleet driving in similar scenarios is maybe an option, but not if that scenario is truly uncommon or dangerous. With Inverted AI's models you can initialize the concrete scenario just prior to the disengagement then simulate a correct solution in the object-list AV 1.0 representation. Simulating sensors using, i.e. NVIDIA Cosmos completes the loop.
Don't have a fleet? Use Inverted AI's vast and ever-growing collection of nearly noise-free log data and APIs to stand up an unparalleled synthetic data generation pipeline.