Program

Agricultural robots must sense and act in environments that are deformable, cluttered, biologically variable, and economically constrained. This work explores embodied sensing and soft actuator strategies that extend robotic perception beyond passive imaging, focusing on two complementary domains: underground soil interrogation using digging robots and above-ground canopy monitoring using soft robotic grippers. Subsurface robots physically enter and manipulate soil to reveal properties that are difficult to infer remotely, including structure, compaction, and root-zone variability. In the canopy, compliant grippers enable gentle interaction with plants, repositioning leaves and stems to improve monitoring while minimizing crop damage. 

Together, these systems highlight both the promise and the adoption challenges of agricultural robotics. Robust hardware must survive field conditions while remaining affordable and maintainable; AI systems must interpret noisy, contact-rich data across changing crops and soils; and deployment must align with grower workflows, policy constraints, and funding models that reward practical value rather than novelty alone. By treating agricultural robots as physical probes as well as autonomous platforms, this work argues for a path from innovation to adoption grounded in reliable embodied interaction, actionable agronomic data, and field-realistic system design.

Despite significant advances in agricultural robotics, post-harvest handling and sorting remain challenging because biological products exhibit large variations in geometry, mechanical properties and ripeness. As a result, robotic systems must combine accurate perception with reliable and gentle physical interaction. Recent advances in soft robotics have significantly improved the safe handling of delicate crops. However, current agricultural robotic systems typically comprise multiple interconnected subsystems, including cameras, external sensors, artificial intelligence, robotic manipulators, and dedicated end-effectors to perceive, evaluate, and manipulate agricultural products. While such modular architecture provides versatility and high performance, it also increases system complexity, calibration requirements, maintenance demands, and deployment costs.

I will discuss an alternative paradigm inspired by biological organisms, in which sensing, actuation, morphology, and thus intelligence, emerge from an integrated physical body rather than from the addition of independent subsystems. In this perspective, for example we have focused on developing optical soft sensing via additive manufacturing, and consequent 3D sensorized structures, or soft monolithic perceptive units (MPU). Building on this, to develop a soft robot, our approach is to co-design sensing, actuation and robot body by linking design and fabrication through simulation. I will suggest this approach by showing the development of a monolithic perceptive gripper that takes inspiration from the elephant trunk tip and that harnesses lattice structures and embedded pneumatic actuation. Then we aim at progressively spanning from adaptive soft grippers to large-scale continuum manipulators inspired by the elephant trunk in which morphology actively contributes to prehensile capability.

This approach is particularly relevant to post-harvest handling and sorting, where the identification of fruit condition, ripeness, damage, and marketability can represent a major source of time and cost. Future agricultural grippers could combine manipulation and assessment during the same physical interaction, using embedded sensing and adaptive morphology to gather information while handling the product. Rather than separating grasping, inspection, and sorting into distinct stages, monolithic robotic systems may support a more integrated process in which physical interaction contributes directly to decision making.

Finally, the opportunities and remaining challenges associated with this direction are discussed, including distributed sensing and actuation, scalable fluidic routing, durability, and the interpretation of different sensory signals. Embedding functionality directly into the robot body offers a promising pathway toward agricultural robotic systems that are simpler, more adaptable, and better suited to the variability of biological products.

Reliable autonomy in unstructured, dynamic environments remains a fundamental challenge for physical AI. While commercial robotic platforms excel in controlled settings, they often fail to generalize to real-world deployments that require robustness, adaptability, and precise control. This talk presents new research directions for resilient autonomy at the intersection of digital twins, physical AI, multi-agent reasoning, and integrated hardware–software co-design that together enable robots to perform delicate and uncertain tasks beyond the capability of off-the-shelf systems. Agricultural environments serve as an ideal stress-test domain for this work: complex, partially observable, highly variable, and safety-critical. Using case studies such as precision robotic manipulation (e.g., grasping deformable biological objects), robot-assisted disease detection, and micro-scale deployment tasks, we demonstrate how foundational advances in perception, planning, and multi-agent AI enable autonomy to operate reliably from controlled laboratory conditions to real-world field environments. The insights translate broadly to other domains requiring robust physical AI, including environmental monitoring, inspection robotics, automated network infrastructure tasks, trustworthy and robust autonomy, and infrastructure-aware digital twins for physical systems. The talk concludes with opportunities for cross-domain collaboration to advance the foundations of resilient, generalizable physical AI. 

Agricultural robots must operate safely and reliably in environments that change dramatically over time. This talk presents semantic mobile-robotics approaches for persistent autonomy in high-value crop production, with applications in in-field logistics, crop inspection, and harvesting support. We combine topological navigation, LiDAR-based stability filtering, and semantic perception to identify persistent environmental structure despite seasonal change, perceptual aliasing, and unreliable GNSS. These methods improve data association and localisation by exploiting stable geometric and semantic landmarks, reducing dependence on high-precision positioning infrastructure. The work demonstrates how robust perception, mapping, and navigation can enable scalable, long-term robot deployment in vineyards and other structured agricultural settings. 

The core idea of this talk is, instead of bringing a sample to a microscope in a lab, can we bring a microscope to a sample in the field? We discuss how we have implemented this concept to enable closing the loop in robotic pollination and reconstructing microscopic 3D scenes. Our solutions leverage modular approaches combining traditional computer vision and robot control techniques in new ways, which inform us about how to proceed into a data-driven future. To this end, we briefly touch on challenges and gaps in control of the robot arms for agriculture.

Artificial intelligence has made remarkable progress in perception, planning, and decision-making, yet deploying AI-enabled robots in agricultural environments remains a significant challenge. Unlike controlled laboratory settings, farms present highly dynamic and unstructured conditions characterized by changing illumination, weather, terrain, crop variability, and limited availability of high-quality training data. As a result, successful agricultural autonomy depends not only on advances in AI algorithms but also on the careful integration of robotic hardware, sensing, and field-ready system design. This talk presents practical lessons learned from developing and deploying autonomous robots for agricultural applications, including perception, manipulation, and field operations. It discusses the challenges of building reliable robotic platforms, collecting and managing data, and designing AI systems across diverse real-world conditions.

 Agriculture is entering a new era where robotics must move beyond automation toward true adaptability. This keynote will explore how next-generation agricultural robots can dynamically interact with farmers, production systems, and complex natural environments by leveraging artificial intelligence and advanced data strategies.

The presentation will highlight how AI-driven perception, learning, and decision-making enable robots to operate reliably in unstructured agricultural settings. It will examine the role of data collection, sharing, and interoperability in improving mission planning, operational efficiency, and strategic farm management. Particular attention will be given to the development of innovative robotic implements and tools designed to support emerging agricultural practices, including diversified, regenerative, and high-value crop systems.

By bridging technological innovation with practical deployment realities, this keynote will provide a forward-looking perspective on how adaptive, intelligent robotics can strengthen farm resilience, productivity, and sustainability, while remaining aligned with the needs of growers and the broader agricultural ecosystem. 

Forestry is an emerging domain for agricultural robotics in which machines must operate on uneven terrain, beneath dense canopy, and under unreliable GNSS. This presentation examines how unmanned ground vehicles, legged robots, aerial systems, and autonomous forestry machines combine LiDAR, cameras, inertial sensing, and onboard intelligence to navigate forests and perform management tasks. Recent field demonstrations in autonomous forest inventory, under-canopy mapping, tree planting, selective harvesting, and material handling show substantial progress beyond remotely operated data collection. However, most systems remain limited to small-scale or supervised trials. The principal barrier to adoption is no longer sensing alone, but the reliable integration of perception, localization, navigation, task execution, and failure of recovery. Using representative real-world deployments, this presentation will identify the platform and sensor combinations showing the greatest operational potential and discuss the technical and deployment advances needed to move forestry robots from experimental prototypes to practical tools for precision forest management. 

Soft robotics offers a promising approach for enabling safe, adaptable, and versatile robotic systems. This talk will present recent advances in the design of modular cable-driven soft robotic manipulators, with an emphasis on reconfigurable architectures that improve workspace scalability and adaptability. The talk will also discuss data-driven control strategies for soft robots, including predictive control and reinforcement learning approaches that leverage system data directly to achieve accurate and computationally efficient control without requiring explicit dynamic models. Together, these advances demonstrate how the integration of innovative soft robotic design and data-driven intelligence can enhance the performance and autonomy of next-generation agricultural robotic systems. 

Recent advances in robotics and artificial intelligence have significantly expanded the capabilities of autonomous agricultural systems. Yet, widespread adoption remains limited by challenges in adaptability, scalability, robustness, and seamless integration into the complexity of real-world farming environments. This talk argues that the next generation of agricultural automation will be driven not simply by autonomous robots, but by Physical AI, embodied intelligent systems capable of perceiving, reasoning, learning, and acting autonomously in dynamic and unstructured agricultural settings.

Drawing on recent research in heterogeneous multi-robot systems, autonomous crop scouting, adaptive field navigation, and AI-enabled agricultural perception, the talk will demonstrate how Physical AI can enable resilient, scalable, and context-aware agricultural operations. Examples from collaborative field robotics and autonomous crop monitoring will illustrate how advances in embodied intelligence, multimodal perception, distributed autonomy, and adaptive decision-making are helping bridge the gap between laboratory research and practical deployment. The talk will conclude by examining the remaining challenges, and by outlining a roadmap for interdisciplinary research that integrates robotics, AI, agricultural sciences, and human-centered system design to accelerate the responsible adoption of intelligent robotic systems for sustainable agriculture. 

Agricultural systems change faster than current monitoring can follow, and disease is one of the clearest examples. Satellite and airborne sensors give broad coverage but limited biological detail, while ground measurements give detail at the cost of scale. The gap between them is filled today by manual scouting and models that are rarely updated once deployed, so the information reaching growers is often out of date before it is used. We describe an approach that closes this loop through autonomous field robotics paired with AI, using grape downy and powdery mildews as representative pathosystems to demonstrate feasibility. Ground robots and aerial systems (e.g., UAS, airborne, and satellite imagery) collect targeted in situ measurements of canopy structure, plant physiology, and disease symptoms across the vineyard. These measurements calibrate and retrain models that translate remote sensing signals into biological quantities as conditions shift across the season. The models in turn direct where robots sample next, concentrating effort where uncertainty is highest. Sampling, inference, and tasking operate as one continuous cycle rather than separate steps through integrated AI and robotics systems.