Purpose & Scope
AI and robotics are transforming agriculture, yet solutions that perform well in demonstrations can struggle when deployed in the field. Limited computing resources, heterogeneous hardware, unreliable connectivity, and imperfect sensor data can all affect system performance and reliability.
Through agrifoodTEF, FBK offers SMEs and startups benchmarking and validation services to assess their technologies under realistic operating conditions, identify limitations, and guide improvements before field deployment.
The presentation will draw on three applications: an orchard system for apple counting and sizing, a vineyard irrigation decision-support solution, and an edge AI system for turfgrass disease detection. These cases illustrate complementary validation approaches, from software performance and stress testing to field-based sensor comparison. For software-intensive applications, FBK has developed a benchmarking toolkit based on open-source tools that enables repeatable experiments on real edge hardware, identifies processing bottlenecks, and reveals reliability risks under increasing workloads.
By sharing practical findings and lessons learned, the presentation will show how evidence-based validation within agrifoodTEF helps companies make better design and deployment decisions.
Learning objectives
- Field-readiness challenges: understand the main factors, such as constrained computing, heterogeneous hardware, unstable connectivity and noisy sensor data, that cause AI and robotics solutions to underperform outside the lab.
- Validation approaches: explore complementary methods for assessing agri-food technologies, including software benchmarking, stress testing and field-based sensor comparison.
- Benchmarking on the edge: learn how repeatable experiments on real edge hardware can expose processing bottlenecks and reliability risks under increasing workloads.
- agrifoodTEF support: discover the benchmarking and validation services FBK offers through agrifoodTEF and how SMEs and startups can access them.
Learning outcomes
- Risk awareness: identify the technical limitations most likely to affect an AI or robotics solution when moving from demonstration to real operating conditions.
- Method selection: choose the validation approach best suited to a given application, whether software-intensive, sensor-based or both.
- Evidence-based decisions: interpret benchmarking and testing results to guide design improvements, hardware choices and deployment strategies.
- Lessons from practice: apply insights from the orchard, vineyard and turfgrass services to anticipate and avoid common pitfalls in your own projects.
Who should attend?
- Agri-tech startups & SMEs
- AI & Robotics Developers
- Product & Technical Managers
- Researchers & Innovation Advisors
AI implementation
Data (data space / bigdata)
Software / AI model