Evidence boundaries in the 2018 Autonomous Greenhouse Challenge
What does the first Wageningen challenge establish, and which questions remain when planning a new commercial pilot?
Scope of this review
An attributed analysis of a published research project. We use the WUR experiment record, the researchers’ later trial account and the original technical discussion. Hamfy had no role in the experiment. This brief contains no original field data or independent financial reanalysis.
How we assessed the evidence
We checked trial scope against the institutional dataset record, responsibilities and the reported outcome against the researchers’ account, and control-design context against the technical paper. We keep those reported facts separate from our proposed transfer and evaluation process.
What the evidence supports
The physical comparison is useful evidence
The documented experiment supports examining an AI strategy in the context of an actual crop and reference operation. It does not settle how the same strategy would perform at another site. The relevant next step is a new, explicitly scoped comparison.
The operating boundary matters
The division of algorithmic and human tasks prevents the result being read as proof of a fully unattended farm. A new project should budget and record all tasks that remain with the operating team.
The objective affects the winning strategy
The reported production and resource-use trade-offs are a reason to write the business objective before deployment. Our proposed acceptance sheet separates output, resource consumption, intervention workload and the economic assumptions used to interpret them.
What this review does not establish
- The experiment concerns one crop, a research facility and a defined period; it is not a multi-site commercial validation.
- Published trial outcomes do not establish a guaranteed improvement for another crop, building or energy tariff.
- We did not download and recompute the raw time series, harvest records or financial scoring, so no new numerical comparison is claimed.
Hamfy’s proposed next step
Use the case to design an auditable pilot: describe the existing controller, the proposed decision layer and the human responsibilities; agree on a baseline and the outcomes that matter; then collect a complete crop-period record. Use the record to decide whether the scope can expand.
Evidence to collect for a real project
- A baseline with crop output, climate, resource use and intervention records from the proposed site.
- A written control boundary with permitted setpoints, overrides and recovery behaviour.
- A comparison plan that addresses crop and environmental differences.
- A transparent outcome report including unsuccessful periods and operating costs.
Source notes
- Primary experiment record · Dataset published 2019Autonomous Greenhouse Challenge: First edition, 2018Wageningen University & Research · Hemming et al.
Establishes the experiment’s location, crop, period, compartments and the categories of available data. Data DOI: 10.4121/uuid:e4987a7b-04dd-4c89-9b18-883aad30ba9a.
- Peer-reviewed trial account · 2020Remote control of greenhouse cucumber production with artificial intelligence: results from the first international autonomous challengeElings et al. · Acta Horticulturae 1294
Supports the division of responsibilities and the qualitative outcome. DOI: 10.17660/ActaHortic.2020.1294.9.
- Peer-reviewed technical paper · 2019Remote Control of Greenhouse Vegetable Production with Artificial IntelligenceHemming et al. · Sensors 19(8), 1807
Describes control approaches and discusses strategy, crop production and resource use. DOI: 10.3390/s19081807.