Truly autonomous vehicles must have the ability to understand and predict the outcomes of complex scenes and scenarios. We believe that achieving safety and reliability in autonomous systems requires cross-disciplinary efforts that involve traditional engineering, and a streamlined development process that incrementally incorporates state of the art methods as they emerge.
After training, our models represent an abstract understanding of the world that is interpretable to us. This abstraction allows us to cope with uncertainty, especially necessary in cases where it is impossible for sensors to “see” their environment due to obstacles. It also allows us to reduce the complexity of predicting the actions of many other actors in the scene. Finally, we predict and plan for worst-case scenarios, so that they do not occur in the real world. We look at our approach as common sense, augmented by artificial intelligence.