VCU-RVIL: A Multi-Sensor Dataset for Benchmark of Simultaneous Localization and Mapping

This paper presents VCU-RVIL-a multi-sensor dataset-for the benchmark of Simultaneous Localization and Mapping (SLAM) algorithms. The dataset contains diverse data sequences captured in various indoor and outdoor environments from a sensor suite and a GNSS-RTK receiver installed on a ground robot (Husky Jackal). The sensor suite consists of a solid-state LiDAR camera, a fisheye stereo camera, a 3D scanning LiDAR, and two IMUs that are popular robotic sensors. It produces synchronized visual, depth, and inertial data. The GNSS-RTK receiver provides ground truth robot positions with centimeter accuracy. The data from the sensor suite and the GNSS-RTK are paired to form the data sequences of the dataset. VCU-RVIL comprises 27 data sequences (size: ~2TB) acquired from various indoor/outdoor environments by driving the ground robot in these environments. The dataset overcomes the drawback of existing SLAM datasets—limited sensor modalities and/or limited operating environments. It can be used by researchers/ developers to support their efforts in developing SLAM methods (e.g., test, evaluate, fine-tune the methods, and compare them with existing SLAM methods), or assist them in deciding the best sensor modality for their SLAM systems. We are currently working to make it publicly available.