February 26, 2026
Calculators out, lasers everywhere
Lidar waveforms are worth 40x128x33 words
Math flex, DIY dreams, and panic about laser traffic jams
TLDR: New research says reading full lidar waveforms with a transformer can see farther in fog and make cleaner 3D maps. Commenters fixated on the title’s math, debated whether hobbyists can actually buy lidar, and worried about laser cross-talk at busy intersections—highlighting real-world cost and safety anxiety.
A fresh ICCV 2025 paper claims a big glow-up for self-driving car lasers: instead of picking a single blip from the returning light, it reads the whole waveform with a transformer (think pattern-spotting AI) to build cleaner 3D maps and see farther—up to 17 meters more in fog. But the community didn’t care about the lab flex; they were busy doing math and making memes. One commenter dropped “168,960,” instantly solving the paper’s cryptic title (40×128×33) and turning the thread into a calculator speedrun. Another cut through the hype with the real question: can regular folks actually buy lidar yet, or is this still billionaire toy territory? Practical curiosity met sticker-shock dread.
Then came the chaos agent: what happens when a dozen robo-cars blast lasers at the same intersection? One user imagined a rush-hour laser rave where nothing can tell which light is whose, warning that all signals should be treated as untrusted—and doubting coders will do the hard work. While the paper’s PDF quietly promises cleaner, longer-range vision, the comments screamed a different headline: numbers are funny, wallets are wary, and the street-level laser drama is just getting started.
Key Points
- •ICCV 2025 paper introduces a learned DSP using a transformer to process full lidar waveforms with neighbor context.
- •Conventional automotive lidar DSPs use per-waveform peak finding, leading to artifacts in low SNR and adverse conditions.
- •Method produces high-fidelity multi-echo point clouds and is trained on synthetic and real data.
- •Evaluated in real-world driving and a weather chamber with conventional automotive lidar.
- •Achieves Chamfer distance improvements (32 cm vs peak finding; 20 cm vs transient imaging) and extends max range up to 17 m in fog and 14 m in nominal conditions.