Is LiDAR necessary | reinvent the wheel #6

or would Elon change his mind?

Ever since engineers started dreaming of cars that drive themselves, they’ve been locked in a high-stakes debate: what kind of “eyes” does a robot car actually need to navigate the complex, unpredictable human world safely?

While many pioneers embraced a multi-modal approach, incorporating cameras, radar, and crucially, LiDAR (Light Detection and Ranging), one prominent player, Tesla, championed a contrarian path: vision-only autonomy, mirroring how humans perceive the world. This divergence has sparked one of the most enduring debates in the field: Is LiDAR truly necessary for an autonomous vehicle to function?

The answer, perhaps unsatisfyingly, isn't a simple yes or no. It hinges critically on the level of autonomy being targeted. Tesla's "Full Self-Driving" (FSD) Beta, operating primarily on cameras, is currently classified as driver-assistance system, requiring constant driver supervision. Conversely, companies like Waymo and Cruise, aiming for truly driverless operation, almost universally rely on LiDAR as a core component of their sensor fusion systems.

Lidar: Painting the world with light

The underlying principle of what LiDAR does isn't new; its roots trace back to the very invention of the laser. The concept emerged shortly after Theodore Maiman demonstrated the first working laser in 1960. Scientists and engineers quickly realized the potential of using focused light pulses for precise distance measurement, analogous to how radar uses radio waves.

Hughes Aircraft Company is often credited with developing some of the earliest LiDAR systems in the mid-1960s, initially exploring laser rangefinders for military applications like tank gunnery and missile guidance. Simultaneously, atmospheric scientists began developing similar systems, coining terms like "lidar" (light detection and ranging) or "laser radar," to probe the atmosphere, measure cloud heights, track aerosols, and monitor pollution. Throughout the 1970s and 80s, LiDAR found broader applications in geodesy, archaeology, and large-scale topographical mapping, typically mounted on aircraft. These early systems, however, were bulky, power-hungry, and prohibitively expensive, far removed from anything practical for a passenger vehicle.

A significant catalyst for automotive LiDAR development came with the DARPA Grand Challenges in the mid-2000s. These autonomous vehicle competitions spurred intense innovation, demanding sensors capable of perceiving the environment for robotic navigation. While still large and costly (often tens of thousands of dollars), the rotating multi-beam LiDAR units developed during this era (like those famously used by early Waymo prototypes) demonstrated the technology's potential for ground vehicles.

This historical trajectory of cost reduction paved the way for the sophisticated automotive LiDAR systems available today. Modern LiDAR for vehicles works by emitting pulses of laser light, often in the near-infrared spectrum (invisible to the human eye). These pulses bounce off surrounding objects, and a highly sensitive sensor measures the precise time it takes for the light to return. By calculating this "time-of-flight" (ToF) and knowing the speed of light, the system determines the exact distance to the reflecting object with remarkable accuracy, often down to the centimeter level. Repeating this process millions of times per second, scanning across a wide field of view, creates a detailed, three-dimensional "point cloud" map of the vehicle's environment, independent of ambient light conditions.

Companies like Hesai Technology have become leaders in automotive LiDAR. Visualizations of their point clouds, such as those generated by sensors like the AT1440, which boasts 1440 laser lines, showcase the technology's power: a rich, geometrically precise representation of cars, pedestrians, curbs, and road debris, rendered in real-time. This direct measurement of distance and shape contrasts sharply with cameras, which infer depth and structure from 2D images.

Hesai’s "world's highest line count" for an automotive-grade LiDAR with 1440 vertical laser lines (compared to 128 in its popular predecessor, the AT128). It boasts an angular resolution down to 0.02 degrees.

Elon Musk's "Doomed" Prophecy and the Tesla Argument Against LiDAR

Elon Musk has been LiDAR's most vocal critic in the automotive space, famously declaring that "LiDAR is a fool's errand" and that anyone relying on it is "doomed" .

Musk's argument, and Tesla's resulting strategy, appears to stem from a core philosophy that permeates much of his engineering thinking: simplify, simplify, simplify. And, relatedly, solve the fundamental problem.

The Bionics Argument: Musk argues from analogy: humans navigate the world primarily using vision (our eyes) and biological neural networks (our brains). Therefore, replicating this system with powerful cameras and sophisticated AI should be sufficient for autonomous driving. Adding LiDAR, in this view, is a "crutch" that prevents solving the harder, but ultimately necessary, problem of computer vision.