Jack’s research interests include; deep learning; multi-agent systems; and computer vision.
- Foster, J., Gudelis, M., Ghalamzan Esfahani, A. (2022) Robotic Perception in Agri-food Manipulation: A Review. arXiv:2208.10580[cs.RO] pre-print.
- ICRA Task-Informed Grasping Workshop – III (2021): Perception in Agri-Food Manipulation: A Review. Watch here.
- AgriFoRwArdS CDT Annual Conference (2022): Uncertainty-Guided Collaboration in Multi-Agent Deep Reinforcement Learning.
- Alan Turing Workshop on Firm-level Supply Network Reconstruction: Variational collaboration: How, when, and from whom to learn in a complex supply chain network.
- Dyson Farming meeting: Towards Robust Continual Learning with Bayesian Adaptive Moment Estimation
- Dyson Farming visit: Uncertainty-Guided Collaboration for Multi-Agent Precision Agriculture
- Corpus Christi Engineering Masterclass: Can AI’s Learn From Each Other?
Other Activities and Outputs
- Member of the AgriFoRwArdS CDT Advisory Board.
- Took part in the AgriFoRwArdS Summer School 2021 resulting in a co-authored presentation at the AgriFoRwArdS CDT Annual Conference 2021: Visual Serving for Human Tracking and Counting (in collaboration with Grzegorz Sochacki, Samuel Wandai Kihara, Kirsten Ayris, Thomas Slattery).
- Co-supervised 6 Mathematics PhD students across UK universities for a research project with the Heilbronn Institute for Mathematical Research.
- NCNR project – robotic sort and segmentation of nuclear material.
Hi I’m Jack, I started out studying Computer Science at Keele before moving to the University of Birmingham to complete my master’s in Robotics. After that, I joined the CDT in 2020. So far, I have really enjoyed working with the cohort, its far more engaging and fun to attend talks or training days with a big group of people you know and are friends with. I’m mainly interested in the autonomy and decision making behind robotics, particularly long-term autonomy, and anything to do with neural nets, genetic algorithms, or multi-agent systems. In my spare time, I enjoy playing the guitar, photography, or nipping into town on the electric scooters dotted around Cambridge!
Lifelong Learning for Sensor-based Precipitation Regression
This project seeks to improve the crop yield of a small farm by optimising the soil-moisture content via precipitation regression. A CNN or RNN, will use wind speed/direction, humidity, temperature, and air pressure to predict future precipitation. Once the network is trained, lifelong learning will be applied to improve the network’s performance when localised within the small farm, using data collected in real-time from a range of sensors and an Arduino. The future precipitation will be combined with the current moisture levels to create predictions on whether the crop requires watering today.
Collaborative Lifelong Learning for Robust Site-Specific Crop Management
Farms are not, in general, homogeneous. As such, the farm-wide treatment and maintenance of crops leads to sub-optimal crop yield or quality. By taking a finer grained approach, crop would receive the necessary care for their needs, rather than the average need of the farm. However, it is impractical for human experts to manually analyse the needs of crop on such a granular scale.
To address this need, we first propose novel machine learning approaches for crop care, such as soil-moisture optimisation via LSTM neural networks. To collect real-time data of the farm a custom sensor system will be developed and several of them deployed to collect relevant environmental data from a small region of the farm.
Finally, to improve the long-term autonomy and overall performance of an agent, a collaborative multi-agent system will be constructed to facilitate the lifelong learning from both an agent’s environment, but also from other agents. This will improve the robustness and performance of agents over long periods of time.