Neuromorphic Vision Applications: From Robotic Foosball to Tracking Space Junk

Event: Western Sydney University Presentation · Duration: 26 min · ▶ Watch on YouTube

Abstract

This presentation by Gregory Cohen from the International Centre for Neuromorphic Systems at Western Sydney University explores diverse applications of neuromorphic vision. He showcases how event-based cameras and neuromorphic hardware are revolutionizing fields from space situational awareness (tracking space junk and satellites) to robotic games like foosball and pinball. The talk emphasizes the advantages of neuromorphic systems in terms of robustness, reliability, power efficiency, and their ability to handle high-speed, dynamic environments where conventional frame-based cameras struggle. Cohen also discusses the development of novel neuromorphic datasets and benchmarks to accurately assess the performance of these cutting-edge systems.

Speakers

  • Gregory Cohen — International Centre for Neuromorphic Systems, Western Sydney University

Talks (1)

  • 00:00:00 — Gregory Cohen: Neuromorphic Vision Applications: From Robotic Foosball to Tracking Space Junk
    • Gregory Cohen introduces the International Centre for Neuromorphic Systems at Western Sydney University and outlines their mission to develop neuromorphic sensors, algorithms, and hardware for real-world applications, including space tracking and robotic games.

Key Takeaways

  • Neuromorphic vision offers significant advantages over conventional cameras for dynamic, high-speed applications due to its event-based nature, leading to reduced data processing, storage, and improved robustness.
  • Astrosite containers, mobile observatories equipped with neuromorphic cameras, are being developed for critical applications like space situational awareness and tracking space junk.
  • Neuromorphic systems excel in scenarios with high motion and variable illumination, enabling tasks like tracking fast-moving satellites or objects in challenging environments.
  • Developing specific, dynamic benchmarks and datasets (like robotic foosball and pinball) is crucial for evaluating and advancing neuromorphic hardware and algorithms beyond traditional frame-based metrics.
  • The ‘tyranny of time’ in real-time applications highlights the need for low-latency processing, a key strength of neuromorphic approaches.

Methods / Models / Datasets Mentioned

  • ATIS camera
  • CCD Sensor
  • Gen4 Sensor
  • Astrometry
  • MNIST
  • N-MNIST
  • ePaper Screen
  • Plane Dropping Dataset
  • Deep Learning
  • Simple Template Matching
  • Two-neuron neural network

Topics

Neuromorphic Vision · Space Situational Awareness · Space Junk Tracking · Robotic Foosball · Robotic Pinball · Event-based Cameras · Neuromorphic Hardware · Adaptive Optics · Benchmarking · Real-time Processing

Notes

Open for commentary — connections to other work, critiques, follow-up reading.