"What they're basically doing is helping the car understand where it is in space and time, and importantly helping the model to understand how it should safely navigate whatever scenario,"
""Clearly that's where you need to bring humans back in," Stone said. "We need to re-hone the dataset, we need to use additional context to retrain the model, deploy your fix, and away you go.""
Robotaxis depend on large quantities of real-world and simulated driving data to train perception and decision systems. Human validators, annotators, and labelers review sensor feeds and mark objects, trajectories, and contextual cues so models can interpret scenes. Labelers identify cones, stop signs, pedestrians, and complex events like police activity or school buses to guide appropriate model responses. Labeled datasets enable teams to re-hone models, add context, retrain, and deploy fixes. Data foundries coordinate contributors and serve clients such as Zoox. The AV-specific workforce that supports these tasks is estimated at fewer than a couple thousand people globally.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]