For autonomous robotic systems to receive widespread adaptions, their exteroceptive sensor systems, navigation, and perception algorithms need to be robust to various operating environment. Thermal cameras have been proposed as a potential supplement exteroceptive sensor to address optical cameras' degradation under low-light or LIDAR's high cost and sparse data products. However, their adoption into localization systems, especially in stereo configuration, has not been sufficiently studied and the amount of datasets featuring thermal sensors for localization and sensor fusion context are noticeably lacking.  The proposed work seeks to acquire a high quality dataset featuring stereo optical, stereo thermal, and micro-electro mechanical system (MEMS) LIDAR alongside multiple inertial navigation systems (INS) enabling navigation research.  For enabling a "ground truth" of the vehicle state, this work will include the GNSS observables enabling high-precision GNSS/INS fusion between three GNSS u-Blox receivers with post process kinematic in a moving baseline configuration and a Honeywell tactical grade inertial measurement unit (IMU) to provide the best solution possible. The planned sequences include traversals in an urban environment as well as a wooded environment. Furthermore, each route will be recorded at three different times, i.e., mid-day, twilight, and before sunrise to provide a spread of environmental conditions with varied temperature profiles, further diversifying the operating environments that could be considered for localization. The dataset will be publicly available both raw data and ROS bags. The resulting dataset aims to provide a means for other researchers to investigate the efficacy of thermal cameras in the context of navigation, especially in unstructured environment.

Pending

• 2025/05/29 - TF bug with thermal camera fixed

• 2024/01/20 - Initial release