Researcher in robotic autonomy, with 3.5 years of experience deploying real robotic systems.

A neuroevolutionary method for automatically designing Dynamic Neural Field architectures. It evolves both the network structure and its parameters, reducing the need to manually design controllers.

The method was evaluated across Dynamic Field Theory tasks and more complex problems requiring intermediate representations to emerge.

Applying neuroevolution to generate bio-inspired controllers for a robot working alongside a person.

The controller evolves internal mechanisms for joint action (through simulation), allowing the robot to interpret its partner’s behaviour and contribute the complementary part of a shared task.

For this particular task, the resulting controller was deployed on a KUKA LBR iiwa using ROS2, vision, and MoveIt2. The best evolved controller was transferred to the physical robot without parameter retuning and tested on object arrangements not seen during evolution.

Best Innovation in HRI — NeuroDesign Award, IEEE RO-MAN 2025

A performance-focused C++20 library and interactive application for building and simulating Dynamic Neural Field architectures in real time.

The simulation engine is designed for efficient execution and is paired with an interactive interface built with Dear ImGui and ImPlot for building, inspecting, and modifying models while they run.

Benchmarked 15–41% faster than the next fastest framework in a cross-platform comparison of four Dynamic Field Theory implementations.

A Dynamic Neural Field architecture for modelling human intent and enabling anticipatory robot behaviour.

The system was built around dnf-composer, integrated with CoppeliaSim for the robotic simulation and an HTC Vive virtual-reality setup for the human participant.

We tested anticipatory and non-anticipatory robot behaviour in a human–robot collaboration study, evaluating both participant perception and task performance.

Anticipatory behaviour improved several collaboration measures and significantly reduced collisions.

A study of robustness in Dynamic Neural Field architectures under progressive neuron loss and synaptic disruption.

The work also investigated recovery through re-learning using Learning from Demonstration (LfD), testing whether damaged architectures could recover behaviour from new demonstrations.

The models continued functioning under substantial degradation, and re-learning recovered behaviour in damaged architectures.

A mobile manipulation system combining a UR10e arm and MiR200 mobile base.

The robot navigates between workstations, docks, picks and palletises parts, and sorts defective parts using machine vision. The system includes shared-workspace safety systems.

The system uses a ROS-based architecture connecting task management, vision, robot control, mobile navigation, and safety monitoring.

A collaborative robot cell developed to automate physically demanding pre-assembly work.

The team analysed the manual process, designed and simulated candidate solutions, built a laboratory prototype, and deployed and certified the final system on the factory floor.

The deployment increased productivity by 7.11%, while reducing ergonomic risk across the evaluated tasks.

A sensor glove controls a five-finger robotic hand over a wireless connection.

An Arduino reads and transmits finger sensor data, while an STM32 processes the data and generates PWM signals to control each finger in real time.