Why LiDAR specifications are not directly comparable

Two values with the same name may come from different targets, detection thresholds and processing. Temperature, sunlight, humidity, frame rate, return count and detection probability affect results. Without a common protocol and raw data, treat a datasheet as a vendor claim, not a benchmark.

A 90% white target returns roughly nine times more light than a 10% dark target in a simplified model. A “200 m” range without reflectivity, target size, sunlight and detection-rate conditions is ambiguous. Compare at least 10% and 90% ranges, then test tires, dark clothing and wet asphalt.

A stated ±2 cm error may be typical, maximum, standard deviation or filtered performance. Separate absolute bias, point noise, repeatability and range resolution. Measure multiple distances on a traceable flat target across temperatures and incidence angles, then report mean, 95th percentile and outliers.

Channels often describe the number of vertical scan lines, not uniform spacing or horizontal resolution. A non-uniform distribution may concentrate beams at the horizon; a non-repeating pattern has no single instantaneous resolution. Compare the largest angular gap in the useful region and the probability of hitting a small object.

Increasing frame rate often reduces points per frame or changes horizontal resolution. Maximum output may include multiple returns, invalid pixels or modes incompatible with maximum FoV. Calculate valid points per square degree per second in your ROI, then measure end-to-end latency and jitter.

IP67 or IP69K describes defined ingress tests, not universal resistance to condensation, salt, mud, solvents, UV, thermal shock or repeated washing. Connectors and mounting can weaken system sealing. Request test reports, derated temperature range, drainage and optical-window cleaning procedures.

Freeze firmware, mode, frame rate, returns, power, synchronization and temperature. Record sensors simultaneously on the same 10% and 90% targets under sun, rain and dynamic scenes. Evaluate range at a chosen threshold, bias, noise, completeness, false points, ROI density, latency, network load and power. Keep PCAP/ROS bags and scripts so the decision is reproducible.