Two religions divide autonomy: LiDAR + sensor fusion vs camera-only vision. Waymo and Cruise swear by LiDAR. Tesla bets on eight cameras and neural nets. Which approach is right for safe autonomy at scale?
The Contenders
📡 LiDAR-Fusion Stack
- Direct 3D point cloud (200m range)
- Works in darkness
- Redundant with radar/camera
- Cost: $500–$1000 per unit (2026)
- Used by Waymo, Volvo, Mercedes Drive Pilot
📷 Vision-Only Stack
- 8 cameras, no LiDAR
- Relies on depth from parallax + AI
- Cheaper BOM, scalable
- Needs massive data & compute
- Championed by Tesla FSD v12+
Weather & Edge Cases
LiDAR excels in direct ranging but struggles in heavy rain/fog where laser scatters. Camera vision struggles in low light and glare but benefits from recognizing semantics — traffic lights, gestures, road signs — that LiDAR alone cannot see. Fusion advocates argue: why choose? Use both.
Cost & Scalability
LiDAR prices have collapsed from $75,000 (2012) to under $500 (2026 solid-state). Yet for a $30k consumer car, even $500 matters at millions of units. Vision-only promises fleet learning from billions of miles of dashcam-like data, but verification is harder.
Our Verdict for 2026
For robotaxis: Fusion wins — safety demands redundancy. For consumer ADAS → autonomy: Vision may win on scale if AI proves robust. The likely future: cheap solid-state LiDAR becomes standard even on vision cars — “vision-centric fusion.”