Workshop on Deep Learning in Agriculture, Forestry and Field Robotics
Deploying mobile robots in unconstrained real-world environments
ECMR25 - Padua, Italy
The workshop aims to provide a forum for researchers and practitioners to exchange ideas, discuss challenges and opportunities, and identify new research directions to bridge the gap between laboratory experiments and real-world deployment in unstructured environments. Moreover, it will foster cross-disciplinary knowledge exchange by bringing together distinguished keynote speakers, workshop presenters, and participants from diverse domains—including robotics, artificial intelligence, computer vision, and human-robot interaction. Industry involvement will further ensure that academic insights are grounded in practical needs, supporting the development of innovative robotic applications with real-world impact.
Topics of interest include, but are not limited to:
Navigation and mapping, comprising algorithms for mapping and localization, path planning, and obstacle avoidance for navigation. Particular emphasis will be given to works tackling long-term autonomy and navigation in uncontrolled environments, under unpredictable and changing working conditions.
Perception and sensor fusion, effective and robust sensory processing is one key prerequisite for the successful deployment of robots in unconstrained environments. The workshop welcomes contributions focused on the challenges of interpreting, integrating, and managing real-world data, especially in the context of operational settings where sensor availability is restricted, e.g., underground and underwater settings, or under plant canopy.
Planning, Reasoning, and Decision-Making, an essential requirement for deploying robots in real-world environments is improving their ability to plan actions and make decisions under uncertainty and changing conditions. We welcome submissions that address various reasoning capabilities, including but not limited to physics-based, spatial, and temporal reasoning. Approaches that enhance the explainability of decision-making processes, particularly in critical applications like search and rescue or infrastructure maintenance, are especially encouraged.
Improved Knowledge Representations, effectively representing the complexity and variability of the real world remains a key challenge in robotics. We invite contributions that propose novel or modular knowledge representations capable of adapting to environmental changes. Submissions may include methods that integrate diverse knowledge types, such as domain-specific, task-specific, commonsense, or factual knowledge, to support robust reasoning and autonomy in the field.
Real-world Datasets and Benchmarks, which capture the high variability and unpredictability of real-world environments. While the availability of diverse datasets is essential, equally important are high-quality annotations and well-defined benchmark tasks. We encourage submissions that not only provide datasets but also include benchmark solutions or baseline results, which help quantify task difficulty and support meaningful comparisons, similar to the approach adopted by benchmarks like KITTI. Such contributions are vital for enabling rigorous, reproducible evaluation of robotic systems in realistic settings.