Embark’s trucks are always learning. Like any self-driving vehicle, they are fed a steady diet of training data —information the trucks can use to learn how to various obstacles it sees. Embark initially managed its training data in-house but in-house but this turned out to be a suboptimal solution.
Not only was labeling obstacles a tedious, expensive, and time-consuming task, it was also prone to error. The team knew that this approach just wouldn’t scale. Instead, they wanted to focus on developing the parts of their business that really mattered: their trucks.