Neuromorphic Computing and Vision for Interactive Robotics

Neuromorphic computing and event-based vision seek to replicate the brain’s sparse, spike-driven information flow, offering a route to robots that see and react with the speed and parsimony of their biological counterparts. This thesis pursues that goal through a single narrative arc that begins with computation, advances through perception, and culminates in adaptive interaction. It starts by turning framebased deep networks into spiking models and deploying them on the SpiNNaker neuromorphic platform. Careful comparisons of conversion algorithms show how timing precision, power consumption, and accuracy interplay, establishing practical guidelines for real-time deployment. With spiking computation in place, the thesis next addresses perception. A dynamic-attention mechanism realised as a recurrent spiking neural network keeps track of multiple moving objects in asynchronous event streams, granting robots the ability to prioritise salient targets while ignoring distractors. An accompanying data pipeline exploits the microsecond resolution of event cameras, supporting both hand-gesture recognition and vibration-based machinery monitoring. In tests, these perception modules maintain high accuracy under challenging lighting and rapid motion, confirming the advantages of event-level sensing. The final stage connects machine perception to human intent. Event-based vision is fused with physiological and behavioural indicators to infer the cognitive load of a human partner, enabling the robot to adjust its actions to the user’s real-time mental state. This multimodal loop closes the gap between low-level spikes and high-level collaboration, demonstrating how neuromorphic methods can underpin more intuitive human–robot teamwork. Taken together, the thesis charts a coherent path from energy-efficient spiking computation, through event-driven attention and sensing, to cognitively aware interaction, illustrating how each layer supports the next in building responsive, robust and scalable neuromorphic robotic systems.