Buckle Up: How Waymo Uses Data to Create a Safer Driving Experience
summary
Waymo claims iterative improvement from 100+ million miles on public roads plus billions of simulated miles, serving in five cities with about a quarter‑million paid rides weekly. The production stack uses sensor fusion from five lidars six radars 29 cameras (sixth‑gen trims cameras to ~13) feeding an encoder‑decoder foundation model (sensor fusion tokens → generative action decoder) augmented by a driving VLM. Simulation vets releases across billions of miles; the Waymo Open Dataset is cited 4,000+ times and used by 53,000+ researchers.
AI-written summary · about
chapters (12)
- 0:00Introduction and Waymo overview
- 3:38Waymo’s mission and global expansion
- 6:04How Waymo collects and uses sensor data
- 10:01Challenges of autonomous driving
- 11:25AI’s role in Waymo’s driving system
- 13:59Waymo’s foundation models for driving
- 16:53The four key data streams at Waymo
- 22:39Waymo’s open data set and challenges
- 26:13Real-world safety scenarios in action
- 30:07Waymo’s safety record and transparency
- 31:31Closing remarks and career opportunities
- 32:12Q&A session
related talks
23mA new way forward for self driving cars
27mDriving Autonomous Vehicles Forward w/Real World Applications
17mWaymo's EMMA: Teaching Cars to Think
31mDeep learning for self-driving vehicle technology
04mInterview
08mdata science in automotive
16mDisrupting Mobility and Building the Autonomy Economy
08mHuman–AI Collab for Reliable Automotive Control Systems