LiDAR vs. Vision-Only: How Self-Driving Cars Actually See

Every self-driving vehicle needs to perceive the world around it. How that perception happens is the most consequential engineering choice in autonomous driving. Two dominant philosophies have emerged: Tesla's vision-only approach using cameras and neural networks, and Waymo's multi-sensor fusion combining LiDAR, radar, and cameras. Understanding both approaches is critical for anyone evaluating the safety and reliability of autonomous vehicles in 2026.

The Vision-Only Approach (Tesla)

Tesla's philosophy is built on a first-principles argument: humans drive with two eyes and a neural network, so a vehicle equipped with cameras and a sufficiently powerful AI should be able to do the same. In 2021, Tesla removed radar from new Model 3 and Model Y vehicles and transitioned to Tesla Vision, an eight-camera system relying entirely on neural networks to interpret raw video data.

• How it works: Eight cameras around the vehicle capture 360-degree video. A deep neural network processes these feeds to detect objects, estimate distances, predict trajectories, and generate driving decisions.
• Key advantage: Low hardware cost. Cameras cost a fraction of what LiDAR units do, making the system scalable across millions of vehicles. Tesla's fleet of consumer cars generates enormous training data.
• Key limitation: Cameras struggle in low-visibility conditions. Tesla's Austin robotaxi service has been forced to shut down in rain, and the system can lose performance in heavy fog, dust storms, and direct sun glare.

The Multi-Sensor Fusion Approach (Waymo)

Waymo takes the opposite approach: redundancy through sensor diversity. The current fleet uses a combination of cameras, LiDAR, and radar sensors that each compensate for the others' weaknesses.

• How it works: Waymo's current vehicles carry 29 cameras, six radar sensors, and five LiDAR sensors. LiDAR fires laser pulses to build a precise 3D point cloud of the environment, measuring exact distances to every object. Radar penetrates rain, fog, and dust. Cameras provide color, texture, and context.
• Key advantage: Redundancy. At Google I/O 2025, Waymo demonstrated its LiDAR detecting a pedestrian in a Phoenix dust storm that was completely invisible on camera. If one sensor type fails, others compensate.
• Key limitation: Higher cost and complexity. Each vehicle requires expensive sensor hardware and intensive calibration. Waymo has partially addressed this: LiDAR units that once cost $75,000 can now be sourced for around $1,000.

Next-Generation Sensor Configurations

Waymo's next-generation platform, expected by late 2026, will streamline its sensor suite to 13 cameras, four LiDARs, and six radars. This reduces hardware cost while maintaining the multi-sensor redundancy philosophy. The partnership with Zeekr for a purpose-built robotaxi platform is designed to bring per-vehicle costs significantly closer to Tesla's.

What the Safety Data Shows

As of early 2026, the safety records tell a clear story:

• Waymo has logged over 127 million fully driverless miles with no safety monitor. Independent research shows an 85% reduction in injury-causing crashes and a 91% reduction in serious-injury crashes compared to human drivers.
• Tesla's Austin robotaxi fleet, which still operates with a human safety monitor in every vehicle, has reported crashes at roughly 4 times the rate of average human drivers, based on Tesla's own mileage data.
• Disengagements: Waymo vehicles average roughly 14,000 miles between disengagements. Tesla's FSD system averages approximately 500 miles.

The Scalability vs. Safety Tradeoff

The core tension remains cost versus safety. Vision-only is cheaper and easier to deploy across millions of consumer vehicles. Multi-sensor fusion delivers stronger safety performance but at higher cost and operational complexity. As LiDAR costs continue to fall, this gap narrows. The market is likely to support both approaches: vision-only for consumer ADAS features, and sensor-fused stacks for commercial robotaxi fleets where safety standards are highest.

The Bottom Line

Neither approach has "won." Tesla's path is cheaper and theoretically more scalable, but draws more scrutiny from regulators and safety researchers. Waymo's sensor-rich approach delivers demonstrably better safety outcomes today, but at a cost that limits deployment to commercial fleet operations. The real question for 2026 is whether vision-only systems can close the safety gap before sensor costs become negligible, or whether redundancy will remain non-negotiable for truly driverless operation.