How Waymo sees through walls

By Stripe

Categories: Product, Startup

Summary

Waymo's AI detects pedestrians hidden behind obstacles by analyzing noisy sensor reflections from movement—demonstrating how redundant sensor data and sophisticated models can predict human behavior before visual confirmation. This reveals a critical competitive advantage in autonomous vehicle safety.

Key Takeaways

1. Peripheral lidar bouncing under obstacles creates noisy but predictive signals; Waymo's AI models extract meaningful patterns from low-quality reflections of pedestrian movement to trigger appropriate vehicle responses.
2. Multi-sensor redundancy enables prediction before confirmation—detecting a pedestrian's feet movement through a bus allows the model to predict their emergence and adjust trajectory preemptively rather than reacting post-detection.
3. Sensor fusion transforms noisy data into actionable intelligence; a single lidar reflection of foot movement becomes sufficient signal for confident pedestrian classification and behavior forecasting in production systems.
4. Real-world edge cases in intersections (red light, cross traffic, occlusion) are where sensor sophistication separates safe from unsafe autonomous systems—this San Francisco incident represents exactly the scenario that breaks naive detection approaches.
5. AI model confidence doesn't require perfect data; detecting foot movement under a bus with enough certainty to predict pedestrian trajectory shows how machine learning extracts signal-to-noise ratios that exceed human-engineered thresholds.

Topics

• Sensor Fusion in Autonomous Vehicles
• Lidar Signal Processing
• Pedestrian Prediction Models
• Occlusion Detection
• Safety-Critical AI Systems

Transcript Excerpt

I know we have pretty darn good sensors. Like it does things that I didn't think it was capable of doing. One one example I can give you, it happened in San Francisco. We're at an intersection. Our light is red. There's you cross traffic. Uh a bus goes by. Light turns green. So we start to go. We're nudging around the bus. And then you see a pedestrian being detected on the other side of the bus. And then you know car responds appropriately. It slows down, goes a little bit wider and then a pede...