The machines that teach
machines to see.

LiDAR stands for Light Detection and Ranging. It works by emitting laser pulses and measuring the time it takes for each pulse to reflect back from objects in the environment. By recording millions of these time-of-flight measurements per second, a LiDAR sensor builds a three-dimensional point cloud — a precise spatial map of everything in its field of view.

Time-of-Flight (ToF) LiDAR sends discrete pulses and measures the round-trip time. FMCW (Frequency-Modulated Continuous Wave) LiDAR sends a continuous beam with a frequency chirp and measures the beat frequency of the returned signal. FMCW provides instantaneous velocity data alongside range and is inherently immune to interference from other LiDAR sensors — a critical advantage in dense traffic.

Geiger-mode LiDAR uses avalanche photodiodes biased above their breakdown voltage, making them sensitive enough to detect individual photons. This enables imaging at extreme ranges — beyond 10 kilometres — and through obscurants like fog, rain, and vegetation canopy. Originally developed for military reconnaissance, the technology is now moving into commercial mapping and autonomous navigation.

Cameras provide rich colour and texture information but cannot directly measure distance. They also struggle in low light, direct sunlight, and adverse weather. LiDAR provides precise 3D geometry regardless of lighting conditions. Most advanced autonomous systems fuse data from cameras, LiDAR, and radar together to create a robust perception stack that compensates for the weaknesses of each individual sensor.

A Vertical-Cavity Surface-Emitting Laser (VCSEL) is a type of semiconductor laser that emits light vertically from its surface. VCSELs can be manufactured in dense arrays on a single wafer, making them ideal for high-volume, low-cost LiDAR illumination. Most 905nm solid-state LiDAR systems use VCSEL arrays as their light source because they can be produced at automotive scale using existing semiconductor fabrication processes.

905nm systems use cheaper silicon-based detectors and mature VCSEL sources, but are limited in power due to eye safety regulations. 1550nm systems can transmit significantly more power (the wavelength is less dangerous to human eyes), enabling longer range, but require more expensive InGaAs-based detectors. The choice depends on the application — automotive tends toward 905nm for cost, while defence and long-range mapping favour 1550nm for performance.

A Single-Photon Avalanche Diode (SPAD) array is a detector that can register individual photons with picosecond timing accuracy. By arranging thousands of SPADs into a focal plane array, you create a sensor that can build a 3D image from an extremely small number of returned photons — enabling LiDAR performance at ranges and in conditions that would be impossible for conventional detectors.

Costs vary dramatically by type. Mechanical spinning units that cost $75,000 a decade ago are now under $1,000. Solid-state flash LiDAR modules for automotive ADAS are targeting $200–500 at volume. Defence-grade Geiger-mode systems still run into the hundreds of thousands. The cost trajectory is downward across every category, driven by semiconductor integration and manufacturing scale.

Sensor fusion is the process of combining data from multiple sensor types — typically cameras, LiDAR, radar, and ultrasonic — into a single coherent perception model. No individual sensor is reliable enough for safety-critical autonomous systems. Fusion algorithms weight each sensor’s strengths and compensate for its weaknesses, producing a perception output that is more robust than any sensor alone.

Real-time LiDAR processing happens on edge hardware inside the vehicle or platform. But training the AI models that interpret point clouds, processing large-scale mapping datasets, and running simulation environments for autonomous vehicle testing all require massive cloud compute. Companies in this space consume significant volumes of GPU compute from providers like AWS, Azure, and Google Cloud.

Princeton Lightwave Inc. was a New Jersey-based manufacturer of InGaAs-based single-photon detectors and Geiger-mode LiDAR systems. The company was acquired by Argo AI (a Ford subsidiary) in 2017 to develop advanced LiDAR for autonomous vehicles. After Argo AI shut down in 2022, elements of the technology were absorbed by Ford and other entities. This publication is not affiliated with the original company — we are an independent review named after the domain.

Expect full solid-state integration — no moving parts, on-chip processing, and sensor-level AI classification running at the detector. FMCW architectures will mature for automotive. Geiger-mode will move further into commercial mapping. Costs will continue to fall as semiconductor foundries dedicate more capacity to photonic integrated circuits. The biggest shift will be from standalone sensors to integrated perception modules that combine emission, detection, and computation in a single package.