Description of the mode of collection:
The data origins from two field campaigns at the two separate sites Märrviken and Björsjö. Märrviken is the first test site and the data were gathered in October 2023. It has very rough terrain with large boulders. Björsjö is the second test site and the data were gathered in September 2024. The terrain of Björsjö is less rough than that of Märrviken but it still includes instances of challenging boulders. The field campaigns were done in close collaboration between Umeå University, SCA forest company, Komatsu Forest, Swedish University of Agricultural Sciences, and Skogforsk.
The terrain data was collected before and during the experiments, using helicopter-borne laser scanning and drone footage-based photogrammetry respectively. The laser scanning data was collected using a Riegl Vux 120 sensor and the target height of the laser scan is 150 meters, resulting in point densities of approximately 1500 points per square meter. The time of scan at the Märrviken site is 25-29 July 2022 and at the Björsjö site 19 August 2023, respectively.
During experiments and regular forwarding work at the Björsjö site, we mounted a 360-camera on top of the forwarder cabin. The 360-camera records video and audio of the forwarder’s surroundings. At the Märrviken site, no 360-camera records were acquired, but, apart from regular video, some key features were recorded with a standard mobile phone camera, and converted into 3D models using the online service from Polycam.
The forwarder was operated by professional forwarder operators employed by SCA. For parts of the experiments of the Björsjö campaign, the operator seat was equipped with a vibration sensor. The vibration sensor is a 3D accelerometer with a sampling frequency of 6000 Hz. For parts of the experiments at Björsjö, we also mounted a number of inertial measurement units (IMUs) on the forwarder wheels, which record angular velocity at 5 Hz. The IMUs are synchronized with the machine data and the 360-video, allowing for detailed analysis of the forwarder’s movement and behavior during the experiments, with specific regard to wheel slip.
The camera data is time-synced using the timestamps provided in the metadata. The IMUs are synchronized by inducing and subsequently identifying characteristic events (rotation about a single axis, brief free-fall) at known time instants at the start and end of measurements. Both synchronization methods achieve a temporal accuracy on the order of one second. To enable post-processing and analysis such as synchronization of data streams and isolation of specific scenarios, start and end times of all experiments and other relevant episodes are manually logged in a field notebook.
Please read the published article and the documentation files included in the dataset for more information.
