LiDAR

Time synchronization is an easily overlooked but critically important challenge in multimodal fusion engineering practice, as different sensors have significantly varying sampling frequencies (LiDAR at 10–20 Hz, RGB cameras at 30–60 fps, IMU at hundreds of Hz).

Traditional LiDAR scanning frequencies are 10–20 Hz, and visual deep learning inference on edge computing devices typically requires 30–80 milliseconds.

In the professional robotics field, SLAM is almost bound to lidar.

The control of such devices is essentially a real-time perception-decision-execution closed loop, requiring fusion of inertial measurement unit, wheel speed encoder, depth camera, or lidar data.

Lidar measures distance by emitting laser beams, but heavy snow, dense fog and icy roads cause beam scattering or abnormal reflection, producing noise or even failure; cameras are susceptible to issues such as snow blocking the lens and lane lines being covered.

Winter temperatures in Toronto can drop below minus 20 degrees Celsius, and heavy snow and icy roads can severely affect the perception capabilities of lidar and cameras.

Typical technical solutions for delivery robots include a combination of multi-line lidar, ultrasonic radar, surround cameras, and RTK differential GPS, running in a fixed road network with high-precision maps.

The mainstream cave surveying method is the compass-clinometer-tape trio of measurements, and modern teams using LiDAR and SLAM technology can achieve centimeter-level accuracy.

An IMU alone is insufficient to handle complex terrain; the more mature approach in the industry is to fuse multiple sensors such as foot-end force sensors, vision or LiDAR, and joint current feedback

The fusion approach of lidar, cameras, and millimeter-wave radar is continuously improving its performance in severe weather, at night, and under complex lighting conditions, reducing the risk caused by perception blind spots.

LiDAR can directly output precise three-dimensional distance point clouds, giving it an advantage in ranging and under harsh conditions such as nighttime and backlighting; high-precision maps pre-store static information such as lane lines, speed limit values and school zone locations.

The proposed solution uses a combination of RGB and infrared cameras as a low-cost alternative, which is more financially feasible compared to expensive sensors such as LiDAR.

The current mainstream technology approach for autonomous trucks uses LiDAR as the primary sensor, fused with cameras and millimeter-wave radar to build a redundant perception layer

As a LiDAR target distance increases, the number of beams that can hit the same target surface decreases with the inverse square of the distance, so when the distance doubles the point cloud density drops to approximately one quarter of the original.

The cost of monocular camera solutions is far lower than that of LiDAR and stereo camera solutions.

Autonomous driving systems rely on three core types of sensors for environmental perception: cameras, LiDAR, and high-definition maps.

LiDAR laser beams are scattered by heavy rain droplets, creating rain-fog noise that severely degrades point cloud quality.

Perception frameworks typically employ multimodal fusion architectures, including RGB-D cameras, LiDAR, and IMU devices, constructing geometric environment maps via SLAM algorithms.

The advantages of pure vision solutions include lower cost, simpler maintenance, and no reliance on mechanically rotating parts prone to damage, but they have inherent limitations under extreme conditions such as strong backlighting, heavy rain, and dense fog.

Tesla differs from competitors such as Waymo, which rely on a technical approach involving LiDAR, high-definition maps, and multi-sensor fusion.