Interview
Dmitri Dolgov: Waymo and the Future of Self-Driving Cars | Lex Fridman Podcast #147
- Autonomous driving is anticipated to define the 21st century as a premier artificial intelligence challenge, with the industry shifting from understanding the problem space to developing driver assist programs and ultimately fully driverless products.
- A strategic pivot toward fully driverless vehicles, identified as occurring around 2013, is viewed as the correct path to transitioning from proof-of-concept demonstrations to scalable deployment services.
- Future fleet operations at increased densities will enable real-time exchange of planning information and dynamic observations to influence vehicle interactions, while the fifth-generation hardware platform is designed for manufacturable large-scale production with viable unit economics.
- Lessons derived from the Phoenix deployment are intended to facilitate the "massive copy and pasting" of systems to new users and locations, with cost and aesthetics not expected to hinder LiDAR deployment due to anticipated manufacturing improvements.
- Technical evolution is predicted to move toward joint fusion and end-to-end models that solve tasks across sensing modalities and temporal levels, aiming to elevate the entire performance curve rather than balancing precision and recall trade-offs.
- Safety objectives focus on building a defensive driver capable of avoiding complex ethical scenarios, while the broader outlook views the accumulation of new experiences and learning as an additive process rather than a replacement of previous stages.
- The industry is expected to continue drawing inspiration from events like the DARPA Urban Challenge to attract talent capable of solving the hardest problems in artificial intelligence.