A Short Introduction to Automotive Lidar Technology

Original Article ↗ Hacker News Discussion ↗

Cameras vs. Lidar for Autonomy

  • Strong debate over camera-only vs. multi-sensor (lidar + radar + cameras) approaches.
  • Some argue that if humans drive with eyes and a brain, two cameras plus enough compute should suffice in theory.
  • Others counter that:
    • Human driving uses multiple senses (hearing, vestibular sense, steering feedback).
    • Human vision and brain are far beyond current automotive cameras/compute.
    • Human performance is actually poor (especially at night; one comment notes night driving is overrepresented in fatalities).
  • Lidar is highlighted as excelling in darkness, low light, fog, and for detecting flat objects (e.g., pallets) that are hard to see optically.
  • Several note that camera-only systems struggle with sun glare, rain, and sunrise/sunset conditions.

Industry Practice and Automation Levels

  • Current production Level 3 systems (e.g., from German and Japanese manufacturers) all use lidar, sometimes multiple units.
  • Chinese OEMs are said to include lidar in many mid‑ to premium‑segment models.
  • Tesla’s system is described as versatile Level 2, not certified Level 3; claimed to need frequent human interventions.
  • Disagreement over how “good” Tesla’s system is:
    • Some report flawless short test drives.
    • Others cite third‑party tests with frequent interventions and insist it’s never safe to look away.

Sensor Fusion, Radar, and System Design

  • Advocates of lidar argue more sensors → better perception; fusion avoids “two systems arguing” by weighting each sensor where it’s strong.
  • Critics worry excess data and complexity can slow or destabilize decision‑making and prefer simpler, camera-only designs.
  • Radar is seen as good for range/“something is there,” but with poor spatial resolution and object discrimination compared to lidar.

Lidar Hardware, Cost, and Form Factor

  • Rotating mechanical lidars remain common due to high range and resolution; flash and MEMS approaches struggle with:
    • Photon starvation and low signal‑to‑noise when illuminating wide areas.
    • Eye‑safety limits on laser power.
    • Limited field of view, steering range, and aperture size.
  • Rotating components are viewed as acceptable in automotive contexts, but corner‑mounted units are criticized as damage‑prone in dense cities and slightly enlarging the vehicle’s effective envelope.
  • Costs have dropped but remain high; some suppliers are exiting; FMCW lidar in particular is noted as technically cool but hard to make cheap for low‑margin automotive markets.
  • Expense is tied to precise optics/electronics and still‑low production volume.

Safety, Regulations, and Health Concerns

  • Automotive lidars are supposed to comply with general laser safety standards (e.g., Class 1).
  • One commenter claims these standards can be “gamed,” and that laser damage thresholds are statistical and tricky.
  • Others argue that:
    • Ratings assume direct continuous viewing; in traffic, exposure per lidar is brief and spread across angles.
    • Solar IR/UV is a larger eye hazard.
  • Long‑term effects of widespread lidar exposure in real driving conditions are described as under‑studied.
  • Anecdote: a high‑power 1550 nm lidar array once damaged a camera sensor at a trade show, raising questions about higher‑power systems.

Reliability, Adversarial Attacks, and Interference

  • Lidar can be blinded by laser pointers or the sun; similar vulnerability exists for human drivers.
  • Some foresee malicious misuse (kids treating it as a harmless prank), but others equate it to already‑serious acts like throwing rocks at cars or shining lasers at pilots.
  • Rotating pulsed lidars with randomized timing are said to handle mutual interference between vehicles better than flash systems.

Consumer and Non‑Automotive Uses

  • Interest in using lidar to scan homes or outdoor scenes at higher resolution than phones.
  • Options mentioned:
    • Professional/industrial handheld and drone‑mounted lidars (thousands of dollars).
    • Cheaper 2D spinning units (e.g., hobbyist devices).
  • Phones and tablets:
    • iPhones and some Android models include depth sensors (ToF/structured light / lidar-like) used with scanning apps.
    • Results are decent for consumer‑grade scanning; photogrammetry remains cheaper for many use cases.