Orchard multisensor dataset for weed and trunk mapping: LiDAR, IMU, RGB-D camera and GNSS recordings under the canopy

Multisensor dataset recorded under the canopy of three orchards in southern Portugal (two orange orchards and one avocado orchard) between April and July 2026, with a purpose-built ground platform. Each session contains synchronised data from a 3D LiDAR, an IMU, an RGB-D camera and a GNSS receiver, stored as ROS 2 bags. This record is the main record of the dataset. It contains the main session, CB-11 (22 minutes, the whole of the Campinas B orchard), the calibration of the platform, a tape survey of 106 trunks that serves as a GNSS-independent reference for evaluating localization and maps, together with the LiDAR map in whose frame the trunk centres are given, and 210 images with labelled weeds and trunks (180 for training and validation and a 30-frame test set). Session CA-10 (Campinas A) is in a linked record, https://doi.org/10.5281/zenodo.23808549, and the other sessions will be released as further linked records. The README lists all eleven sessions. To protect the location of the private orchards, the GNSS fixes in the recordings were moved to a synthetic origin at latitude 0, longitude 0: their timing and relative geometry are preserved, but not the absolute position. Each session also includes the track in local East–North–Up coordinates. This dataset accompanies the MSc thesis “Mobile Robot Navigation for Vegetation Mapping in Agricultural Applications” (Instituto Superior Técnico, Universidade de Lisboa, 2026).