Warehouse Operations Beat by an Open Source Package

Last month, I was squatting in the warehouse debugging SLAM for an AGV when I hit a pitfall I've fallen into countless times before. At the corner of the shelf aisle, the point cloud from the LiDAR always came out with a blurry blob, and the visual odometry drifted straight off course whenever the lighting changed. It took me three days of manually collecting data, calibrating, and re-running tests just to barely squeeze the localization error down to under 5 centimeters. Afterwards, I thought: if there were a ready-made dataset with real-world physical interaction noise, where I could tweak parameters and validate algorithms without building everything from scratch and burning cash, wouldn't that be great?