Autonomous Vehicle Technology: Sensors, SAE Levels, and the Safety Debate

Waymo's Robotaxis Have Driven Over 20 Million Miles Without a Human Driver

As of early 2024, Waymo — the autonomous vehicle subsidiary of Alphabet — had accumulated over 20 million fully autonomous miles (with no human driver) on public roads, primarily in Phoenix, San Francisco, and Los Angeles. Its commercial robotaxi service, Waymo One, completed over 700,000 paid passenger trips in the fourth quarter of 2023 alone. A peer-reviewed study published in Nature Communications in December 2023, analyzing 7.1 million Waymo miles in San Francisco and Phoenix compared to human driving in the same areas, found Waymo had 6.8 times fewer injury-causing crashes and 2.3 times fewer police-reportable crashes than human drivers. Autonomous vehicle technology has moved from research demonstration to a measurable safety argument — at least in the limited operational design domains where Waymo currently operates.

Autonomous vehicles use a combination of sensors, artificial intelligence, high-definition maps, and vehicle control systems to navigate, detect obstacles, and make driving decisions without human input. The sector encompasses passenger vehicles, commercial trucks, robotaxis, and agricultural and industrial machinery.

SAE Automation Levels

The Society of Automotive Engineers (SAE) J3016 standard defines six levels of driving automation, from no automation to full automation. These levels define who or what is responsible for driving tasks:

The Sensor Suite

Autonomous vehicles perceive the world through complementary sensor modalities, each with distinct strengths and weaknesses:

• LiDAR (Light Detection and Ranging): Pulsed laser beams measure distance by time-of-flight, generating a precise 3D point cloud of the environment. High accuracy at range; excellent for detecting object shape and size. Limitations: performance degrades in heavy rain or fog; historically expensive (early Velodyne HDL-64 units cost $75,000). Solid-state and MEMS-based designs have brought costs below $500. Waymo uses LiDAR as a primary sensor; Tesla has explicitly rejected it.
• Radar: Millimeter-wave radar (76–81 GHz) measures distance and velocity using Doppler effect. Penetrates rain, snow, and fog well. Lower spatial resolution than LiDAR. All commercial L2+ systems include radar for long-range vehicle detection and adaptive cruise control.
• Cameras: Provide color, texture, and semantic information — critical for reading traffic signs, lane markings, and traffic light states. High resolution; low cost. Require substantial AI processing. Tesla's Full Self-Driving system uses cameras exclusively (camera-only architecture).
• Ultrasonic sensors: Short-range obstacle detection for parking and low-speed maneuvering.

The Perception and Decision Pipeline

Raw sensor data is processed through a multi-stage software pipeline:

• Sensor fusion: Data from all sensors is combined into a unified environmental model (the "world model"). Kalman filters and deep learning-based fusion algorithms are commonly used.
• Object detection and classification: Convolutional neural networks identify vehicles, pedestrians, cyclists, lane markings, traffic signals, and road boundaries in the sensor data.
• Prediction: Machine learning models forecast how detected objects (especially other vehicles and pedestrians) will move over the next few seconds.
• Planning and decision-making: Path planning algorithms (A*, Monte Carlo Tree Search, or learned planners) select the vehicle's trajectory given the world model and traffic rules.
• Control: PID controllers or model predictive control systems translate the planned trajectory into steering, throttle, and brake commands.

Safety Debate: Statistics vs. Edge Cases

The NHTSA Standing General Order requires reporting of crashes involving ADAS or autonomous driving systems. Through early 2024, Tesla vehicles accounted for the majority of reports due to fleet size (Tesla has the most Level 2 ADAS-equipped vehicles on US roads). GM Cruise's Level 4 robotaxi service was suspended in October 2023 after a pedestrian collision in San Francisco in which the vehicle's response after the initial impact worsened injuries — highlighting that Level 4 systems can fail in specific scenarios despite impressive average performance statistics.

• Waymo's published safety data uses "adjusted crash rates" compared to human drivers in equivalent conditions — the most methodologically rigorous comparison yet published for a commercial AV service.
• The "long tail" problem — rare, novel scenarios not well-represented in training data — remains the central unsolved challenge for reaching Level 5 capability.
• V2X (vehicle-to-everything) communication, enabling vehicles to share sensor data with each other and with infrastructure, is widely expected to improve AV performance by extending effective sensor range beyond what onboard sensors alone can achieve.

Federal AV regulation in the United States has lagged technological development, with NHTSA's AV guidance documents remaining non-binding. California's DMV and several other states have developed the most detailed state-level regulatory frameworks. The EU's ALKS (Automated Lane Keeping Systems) regulation, in force since 2021, provides the first legally binding performance requirements for Level 3 systems in any major market.