End-to-end neuromorphic navigation for complex MAV environments

Establish a neuromorphic navigation paradigm for micro aerial vehicles that integrates event-based sensing, spike-based computation, and neuromorphic hardware deployment for operation in real-world complex environments.

Background

The paper identifies a gap between existing neuromorphic navigation studies and a fully integrated autonomous navigation system. Prior work commonly introduces neuromorphic components only at the sensing stage or only at the computation and decision-making stage, rather than combining event cameras, spiking neural networks, and neuromorphic processors in a unified closed-loop framework.

The unresolved challenge concerns deploying such an integrated system on micro aerial vehicles operating in complex real-world environments. The authors note that immature hardware and software ecosystems, the lack of explicit structural information in event streams, limitations in efficient spiking representations and end-to-end training, and restricted neuron models and operators on neuromorphic processors have kept prior systems focused mainly on simpler subtasks such as pose estimation or motion control.