Neuro-inspired spike-based motion: from dynamic vision sensor to robot motor open-loop control through spike-VITE

Area of Science:

• Robotics
• Neuroscience
• Computer Engineering

Background:

• Current robotic systems often lack biological realism in movement control.
• Bridging the gap between bio-inspired sensors and robotic actuators requires efficient real-time data processing.

Purpose of the Study:

• To present a complete spike-based architecture for robotic control, mimicking human movement from a biological perspective.
• To demonstrate real-time spike processing for seamless data flow between sensors and actuators.

Main Methods:

• Developed a layered architecture: Dynamic Vision Sensor (retina), visual processing, neuro-inspired trajectory generation (SVITE), and actuation (PFM).
• Implemented a spike-processing strategy and utilized the Address Event Representation (AER) protocol for inter-layer communication.
• Integrated the open-loop controller onto an FPGA using AER interfaces.

Main Results:

• Successfully demonstrated a viable spike-based controller for robotic platforms.
• Achieved control of robots with a high number of Degrees of Freedom (DoF) using minimal hardware resources (2% of Xilinx Spartan 6).
• Showcased low power consumption (3.4 W) for the developed system.

Conclusions:

• The presented spike-based architecture effectively reproduces intended movements with biological features.
• The Address Event Representation (AER) protocol is suitable for communication between processing and actuation layers in spike-based systems.
• The system offers significant advantages in terms of hardware efficiency and power consumption for complex robotic control.