An FPGA hardware/software co-design towards evolvable spiking neural networks for robotics application

Area of Science:

• Neuroscience
• Computer Engineering
• Robotics

Background:

• Spiking Neural Networks (SNNs) offer bio-inspired computational models.
• Hardware acceleration is crucial for real-time SNN applications.
• Evolving SNNs presents unique implementation challenges.

Purpose of the Study:

• To present an effective hardware realization of a novel Evolvable Spiking Neural Network (ESNN) on Field Programmable Gate Arrays (FPGAs).
• To leverage FPGA advancements for a flexible ESNN hardware/software co-design.
• To demonstrate the ESNN's capability in embedded intelligent systems for robotics.

Main Methods:

• Developed a hybrid learning algorithm combining Spike Timing Dependent Plasticity (STDP) and Genetic Algorithms (GA).
• Utilized a partitioned hardware/software co-design approach on FPGAs.
• Implemented the ESNN as an embedded controller for a robotic system.

Main Results:

• Achieved effective hardware realization of the ESNN paradigm on FPGAs.
• Maximized FPGA flexibility for the ESNN implementation.
• Successfully applied the ESNN for autonomous navigation and obstacle avoidance in a robotic controller.

Conclusions:

• The proposed approach enables efficient hardware implementation of ESNNs on FPGAs.
• The hybrid STDP-GA learning is effective for complex tasks.
• ESNNs are viable for embedded intelligent systems in robotics, particularly for navigation and obstacle avoidance.