Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Robotic Arm Control with a Spiking Recurrent Neural Network on a Digital Accelerator

Alejandro Linares-Barranco, Luciano Prono, Robert Lengenstein, Giacomo Indiveri, Charlotte Frenkel

TL;DR

This paper addresses online adaptive control using spiking recurrent neural networks on edge hardware. It adapts the ReckOn RSNN accelerator to a Xilinx MPSoC (Pynq-ZU) and provides a Python-based interface for embedded training and inference, validated on adaptive robotic arm control. The approach achieves high data throughput with $3.8\times 10^6$ events/s and competitive accuracy ($88.9\%$ training, $83.3\%$ test) on a lemniscate-trajectory task, while using a moderate portion of programmable logic resources. The work demonstrates the practicality of fully digital, open-source RSNN accelerators for low-power, low-latency edge neuromorphic applications in robotics.

Abstract

With the rise of artificial intelligence, neural network simulations of biological neuron models are being explored to reduce the footprint of learning and inference in resource-constrained task scenarios. A mainstream type of such networks are spiking neural networks (SNNs) based on simplified Integrate and Fire models for which several hardware accelerators have emerged. Among them, the ReckOn chip was introduced as a recurrent SNN allowing for both online training and execution of tasks based on arbitrary sensory modalities, demonstrated for vision, audition, and navigation. As a fully digital and open-source chip, we adapted ReckOn to be implemented on a Xilinx Multiprocessor System on Chip system (MPSoC), facilitating its deployment in embedded systems and increasing the setup flexibility. We present an overview of the system, and a Python framework to use it on a Pynq ZU platform. We validate the architecture and implementation in the new scenario of robotic arm control, and show how the simulated accuracy is preserved with a peak performance of 3.8M events processed per second.

Adaptive Robotic Arm Control with a Spiking Recurrent Neural Network on a Digital Accelerator

TL;DR

This paper addresses online adaptive control using spiking recurrent neural networks on edge hardware. It adapts the ReckOn RSNN accelerator to a Xilinx MPSoC (Pynq-ZU) and provides a Python-based interface for embedded training and inference, validated on adaptive robotic arm control. The approach achieves high data throughput with events/s and competitive accuracy ( training, test) on a lemniscate-trajectory task, while using a moderate portion of programmable logic resources. The work demonstrates the practicality of fully digital, open-source RSNN accelerators for low-power, low-latency edge neuromorphic applications in robotics.

Abstract

With the rise of artificial intelligence, neural network simulations of biological neuron models are being explored to reduce the footprint of learning and inference in resource-constrained task scenarios. A mainstream type of such networks are spiking neural networks (SNNs) based on simplified Integrate and Fire models for which several hardware accelerators have emerged. Among them, the ReckOn chip was introduced as a recurrent SNN allowing for both online training and execution of tasks based on arbitrary sensory modalities, demonstrated for vision, audition, and navigation. As a fully digital and open-source chip, we adapted ReckOn to be implemented on a Xilinx Multiprocessor System on Chip system (MPSoC), facilitating its deployment in embedded systems and increasing the setup flexibility. We present an overview of the system, and a Python framework to use it on a Pynq ZU platform. We validate the architecture and implementation in the new scenario of robotic arm control, and show how the simulated accuracy is preserved with a peak performance of 3.8M events processed per second.
Paper Structure (5 sections, 5 figures)

This paper contains 5 sections, 5 figures.

Table of Contents

  1. Introduction
  2. The RSNN accelerator ReckOn
  3. MPSoC implementation
  4. Experimental Results
  5. Conclusions

Figures (5)

  • Figure 1: Block diagram of the ReckOn accelerator (simplified from reckon_frenkel_2022).
  • Figure 2: Left: Block diagram of ReckOn on the Pynq-ZU board. Dark blue blocks are implemented in the PL to support ReckON, while light blue blocks correspond to the PS and IPs from Xilinx library for connectivity. Right: RSNN used in the experiments.
  • Figure 3: ReckOn test flow diagram
  • Figure 4: Left: ED-Scorbot SPID controllers and spiking activity recording scenario for the collected dataset. Right: Accuracy on tests during hyperparameter search.
  • Figure 5: Accuracy in train/test for best hyperparameters.