Robot-executed longitudinal sloshing dataset for rectangular tank liquid handling

This dataset supports the manuscript Vision-Based Measurement and Large-Scale Experimental Validation of an Analytical Model for Longitudinal Sloshing in Rectangular Tanks for Robotic Liquid Handling.

The dataset contains robot-executed longitudinal transport trajectories and the corresponding experimentally measured liquid response used for the large-scale validation of an equivalent mass-spring-damper (MSD) model for longitudinal sloshing in a rectangular tank.

The experimental setup consists of a T75 cell culture flask moved by an industrial manipulator. The liquid response was measured with a compact, non-intrusive, single-camera side-view vision-based measurement system. The measured output is the longitudinal wall-wave elevation, consistently with the output variable adopted in the analytical MSD model.

The dataset includes 9822 valid trajectories. The first 5000 trajectories are standard trajectories generated using motion laws commonly adopted in robotics and industrial automation, including polynomial splines, trapezoidal and modified trapezoidal profiles, harmonic motion and cycloidal motion. The remaining 4822 trajectories are optimized trajectories generated by exploiting the equivalent MSD model to reduce sloshing while preserving a smooth acceleration profile.

The main data files are:

INPUT_TRAIN.npy: imposed longitudinal acceleration trajectories of the container, in m/s2;

OUTPUT_TRAIN.npy: vision-based measured longitudinal wall-wave elevation, in mm;

trajectory_metadata.csv: per-trajectory labels, subset information and summary quantities;

time_vector.csv: reconstructed sample index and time vector;

load_dataset.py: minimal Python script to load, inspect and plot the dataset.

The arrays have shape (9822, 1500). Each row corresponds to one robot-executed trajectory, and the same row index links the input acceleration and the measured wall-wave response. The validation interval used in the manuscript corresponds to the first 750 samples, i.e. 3 s, including both the active motion phase and the initial free-decay response.

The dataset is intended to support reproducibility and future benchmarking of sloshing measurement and prediction methods under realistic robotic operating conditions.