Scale Transform 2021 - Taking Autonomous Driving from Research to Reality with Drago Anguelov

Autonomous driving research currently centers on overcoming the limitations of deep learning in handling long-tail, rare events and refining intermediate representations for behavior prediction and planning. Drago Anguelov, Head of Research at Waymo, emphasizes that while perception is largely solved for common scenarios, the industry must transition toward models that exhibit greater robustness and handle multi-modal human interactions. Effective behavior prediction requires moving beyond independent trajectory guessing to joint prediction and planning, supported by more stringent metrics that account for complex agent interactions. Furthermore, high-fidelity simulation acts as a vital data multiplier, allowing for the safe, scalable evaluation of diverse scenarios and the mitigation of human error, which causes the vast majority of real-world accidents. By leveraging these advanced simulation frameworks and structured data representations, the industry can accelerate the deployment of safe, fully autonomous systems.