Exwayz’s Post

Two months back, when we were in Vegas for the #CES2024, we deployed our own #dual sensor mapping setup. As shown in the photo below, we put one sensor horizontally, and the second one with a strong 60° tilt. One sensor alone already measures high quality and abundant data, thus why using an additionnal one? 🤔 Actually, the sensor mounted horizontally covers decently the surroundings of the vehicle, which is a good property for doing LiDAR #slam, while the tilted sensor allows to measure points that are high on the closeby structures, and directly adjacent to the vehicle. One could think that using the tilted sensor alone would be sufficient, as it gathers the most valuable data for mapping purposes, however, the #LiDAR SLAM can get lost with a very tilted configuration because of the lack of features measured by the sensor (even Exwayz cannot do magic when there is not enough geometry 🔮). We already discussed that in an older post : https://lnkd.in/gZvC6trP Doing sensor fusion is the key to get the best of the two worlds 💡 The idea is to fuse the two point cloud feeds to create "virtual frames" composed of both LiDAR measures and to use this as the input data for the #SLAM. However, doing this LiDAR #fusion requires 1) excellent time synchronization 2) phase locking the sensor 3) super accurate extrinsic calibration. The two first points are mainly hardware challenges (thanks Hassan 🤖 Bouchiba and Antoine Plat for dealing with that 😅) that require playing around with the sensors setup and network configuration. Having the time synchronization and phase locking is necessary to know how to compose the frames of each sensor, based on the time of each point. Regarding the sensor extrinsic calibration, it can be delt with in multiple ways, but it is worthy noticing that a tiny 0.5° error in the #calibration between the two sensors can already lead to 40cm offsets between points measured at 50m. That’s why being able to compute this #calibration super accurately is critical for the overall quality of the point cloud. Fortunately, Exwayz SLAM technologies and clever ideas allow to compute LiDAR - LiDAR calibration in a software fashion that is able to reach such lever of precision 📐 Once these challenges are overcome, we can create centimer-accurate, exhaustive, and reliable 3D point clouds with a processing time that is 5x shorter than the acquisition duration! Ready to map cities! 🏙️ #LiDAR #SLAM #dual #sensor #setup Aubin Tchoi