Approaching Retinal Ganglion Cell Modeling and FPGA Implementation for Robotics.

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Robotics

Background:

• Neuromorphic engineering applies biological neural computation principles to engineering challenges.
• Retinal Ganglion Cells (RGCs) in the retina detect approaching objects by firing action potentials.
• Dynamic Vision Sensors (DVS) provide event-based visual input, enabling end-to-end event-driven processing.

Purpose of the Study:

• To mimic the approaching object detection functionality of RGCs for robotic applications.
• To compare software and hardware (FPGA) implementations of this RGC-inspired function.
• To evaluate the hardware latency and reaction time for robotic attention/reaction mechanisms.

Main Methods:

• Developed a software model in Java for RGC approaching detection.
• Implemented the RGC function on a Spartan 6 FPGA for hardware acceleration.
• Utilized an asynchronous event-driven Dynamic Vision Sensor for visual input.
• Measured processing time per event and robot reaction latency.

Main Results:

• Software model achieved an average processing time of 370 ns/event.
• Hardware (FPGA) implementation reduced processing time to 160 ns/event at 50 MHz.
• The system exhibits non-deterministic responses due to bio-inspired characteristics.
• A mobile robot demonstrated a reaction time of 90 ms to approaching objects.

Conclusions:

• FPGA implementation of RGC-inspired approaching detection significantly reduces latency compared to software.
• The developed system provides an effective attention/reaction mechanism for robots.
• The reaction time is competitive with existing event-based robotic systems and human visual reaction times.