johansson-delayed-sensor-control

Johansson four-tank trajectories with delayed sensors

Simulated trajectories of the Johansson quadruple-tank benchmark, for testing estimators and controllers when the lower-tank sensors arrive one sample late. There are 314,496 rows across four splits. These are simulations, not plant data.

from datasets import load_dataset
ds = load_dataset("sankalpsthakur/johansson-delayed-sensor-control", "default")

Keep these roles. Don't mix rows from shared trajectories at random.

Which columns a controller may use

• At decision k, the allowed inputs are the public nominal regime, the input_* columns, and earlier rows of the same episode.
• Condition names, seeds, and all result_* and label_* columns are never inputs. An applied result_* voltage becomes history only after it is applied.
• label_next_measured_lower* is y[k+1], which only arrives at decision k+2 because of the delay.
• True tank levels, sensor bias and flow labels are for scoring only.
• Never build history across episode_id boundaries.

Missing observations are empty fields. Warmup intervals are left out here and kept in the full research kit. Units and timing cutoffs are in schema.json; split counts and hashes are in manifest.json.

Limits

The confirmation conditions and seeds are public in this release, so a new claim needs a fresh frozen evaluation.

Source

Report, equations, selected models and full research kit. The source benchmark is Johansson, IEEE TCST 2000. These trajectories are independently generated simulations.

The previous card is in HISTORY.md, and the original export card in CLOUD-CARD.md.