Installing a LiDAR on an AGV, from the point cloud to the stable output of obstacle avoidance signals, there are more pitfalls than imagined.

I am responsible for sensor scheme selection in a company that specializes in warehouse automation. Last year, I started to replace the perception scheme for the AGVs in the factory. Before, we used a 2D single-line LiDAR that could only scan a single plane, and the tray feet under the shelves and the scattered objects on the ground were often not detected. After switching to a 3D radar, the situation improved significantly. Today, I will summarize this experience.

The radar uses a domestic M360 3D LiDAR from TanTu ZhiXing. The main reasons for choosing it will be discussed later.

Installation and Calibration

Let's talk about installation first. The M360 has a size of 78×78×81mm and a weight of 408g, not too large. We install it directly below the front of the AGV, with an inclination angle of about 15° downwards. Why downwards? Because the vertical FOV is 70°, from -10° to +60°, the inclined installation covers both the ground and the front shelves. With another radar that had a vertical FOV of 59° before, the same installation angle resulted in much less ground coverage.

The first thing to do after installation is to calibrate the coordinate system. There is a very convenient feature in the M360: it is built-in with a 6-axis IMU and supports PTP v2 time synchronization. The timestamp of the point cloud data and the IMU data is synchronized at the nanosecond level. With other radars, it was necessary to attach an external IMU module and then manually synchronize the timestamps, which was very cumbersome. The M360 directly saves this step, ensuring data consistency during coordinate transformation.

Point Cloud Processing

Calibration and Point Cloud Processing

Obstacle Avoidance Configuration in Practice

For obstacle avoidance, we use the DWA algorithm. The specific method is to downsample the point cloud from the M360, perform obstacle expansion within a 3-meter range around the robot, and then pass it to the DWA scoring function. There are several details worth mentioning.

The first is the blind zone issue.The near blind zone of the M360 is only 5cm. This data is smaller than most 3D radars. In the storage scenario, the ground is often filled with bolts, cables, and tray feet, which are about 2~5cm in height. With the radar that had a blind zone of more than 10cm before, these small obstacles were often missed, and the AGV ran over them several times. After switching to the M360, ground obstacles beyond 5cm are basically detectable.

The second is rain, fog, and dust.A section of our warehouse is adjacent to the shipping outlet, and the door is sometimes left open during loading and unloading, allowing water mist to enter from the outside. With the previous radar, the noise points were notably more when this occurred, and the point cloud was filled with scattered noise points. The M360 supports dual echo mode, which can separately receive forward and backward echoes, penetrating raindrops to see objects behind. After switching to dual echo, the noise points in the rain decreased significantly. It also features a rain and fog detection function that automatically determines whether there is precipitation in the environment and then adjusts the point cloud output strategy.

The third is IP protection.This matter is a bit embarrassing. The radar we used before did not have an IP rating, and we didn't pay much attention to it. Until one time, when the AGV passed through a ground recently cleaned by a cleaner, some water splashed on the base, and the radar reported an error after working for a few minutes. The M360 is IP67, completely dustproof, and can withstand short-term immersion. After installation, we no longer have to worry about damp and dusty areas in the warehouse.

Power Consumption and Battery Life

Finally, let's talk about power consumption. The working power consumption of the M360 is less than 4.5W. Our AGV uses a 48V lithium battery pack, and the previous radar was about 8W. The 3.5W difference may not seem significant, but AGVs usually need to run continuously for more than 8 hours, and with two shifts a day, the power difference has an impact on battery life and heat dissipation design. Low power consumption means that smaller heat sinks can be used, and the overall structure can be more compact.

Feelings after using it for more than a year

This M360 has been running in the warehouse for more than half a year, and there have been no major problems with AGV obstacle avoidance. Two situations that used to happen have been improved: ground small obstacles are no longer missed, and noise points in the rain are controlled. The built-in IMU saves the trouble of external modules, and the IP67 protection level is also reassuring in the warehouse environment.

If you are also involved in sensor selection for AGVs, the M360's 5cm near blind zone, dual echo penetration through rain and fog, IP67 protection, and built-in IMU with PTP synchronization are indeed very useful in storage scenarios.

Data Source: M360 product manual (TanTu ZhiXing) and actual deployment experience. The content of this article is for reference only, and the actual performance should be based on the official latest parameters.