Reproduce “A compact neuromorphic system for ultra-energy-efficient, on-device robot localization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Adam D. Hines, Michael Milford, Tobias Fischer
A fully neuromorphic system called Locational Encoding with Neuromorphic Systems (LENS) enables ultra-energy-efficient, on-device robot localization by integrating a Dynamic Vision Sensor (DVS) and a compact Spiking Neural Network (SNN) on a specialized neuromorphic chip. This system consumes only 2.7 mW, utilizes models under 180 KB, and achieved competitive localization accuracy with a Recall@1 of 0.88 on an 8 km dataset and 0.64 for on-robot indoor navigation.
Key insights
- Hardware-algorithm co-design, specifically integrating DVS and a neuromorphic processor with a compact SNN, is essential for achieving ultra-energy-efficient VPR on edge devices.
- A simple, three-layer SNN, trained with a combination of unsupervised STDP and supervised delta learning, can achieve competitive VPR accuracy while maintaining extreme model compactness.
- Direct processing of raw, asynchronous event streams from DVS sensors on-chip maximizes the benefits of event-driven vision, leading to reduced data redundancy and processing overhead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.