Foundation Models for Agricultural Robotics and Field Perception

Project Description

This internship explores the development and evaluation of foundation models for agricultural robotics applications. The project will investigate how large, pre-trained models (e.g. vision, multimodal, or representation-learning models) can be adapted and fine-tuned to support perception and understanding in complex agricultural environments.

The intern will work on designing a learning pipeline that leverages diverse agricultural data (e.g. images, spatial maps, or temporal observations) to build generalisable representations that can be reused across tasks such as crop monitoring, anomaly detection, or field mapping. The emphasis will be on transferability, data efficiency, and robustness in outdoor settings
.

Motivation

Agricultural robotics suffers from fragmented, task-specific models that do not generalise well across crops, seasons, or sensing conditions. Foundation models offer a promising route to overcome these limitations by learning rich, reusable representations from large and heterogeneous datasets. This project contributes to ongoing L-CAS research on scalable autonomy and aims to bridge modern foundation-model research with real-world agricultural robotics challenges.

• Strong programming skills in Python
• Background in machine learning, robotics, computer vision, or AI
• Familiarity with Linux-based development

• Experience with deep learning frameworks (e.g. PyTorch)
• Exposure to self-supervised or representation learning
• Basic knowledge of robotics or spatial data (ROS, maps, sensors)

• Hands-on experience developing and adapting foundation models
• Understanding of self-supervised and transfer learning in robotics contexts
• Working with large-scale, real-world agricultural datasets
• Evaluating generalisation and robustness in outdoor environments
• Research skills aligned with cutting-edge AI and robotics publications

• The project scope can be tailored to BSc, MSc, or summer research internships
• The work may contribute to ongoing L-CAS agri-robotics and digital agriculture projects
• Opportunities may exist for co-authorship on research outputs or follow-on PhD study