基于FPGA的移动机器人SNNs走廊场景分类器

Mobile Robots’SNNs Corridor-scene-classifier Based on FPGA

神经形态芯片是类脑计算的重要研究内容之一,神经网络的硬件实现是神经形态芯片实现的基础。具有生物似真性的脉冲神经网络(Spiking Neural Networks, SNNs),通过尖脉冲(Spikes)传递时空信息,更适于用硬件实现,是实现类脑计算的主要工具之一。该文提出一种基于FPGA的移动机器人SNNs走廊场景分类器:将移动机器人超声传感器信息进行脉冲编码后输入到SNNs走廊场景分类器中,通过FPGA分类器的脉冲输出模式来判断机器人所处的走廊场景,从而提高机器人的环境感知能力和自主性。详细讨论了脉冲积分点火神经元模型的FPGA实现原理,以及基于此神经元模型的SNNs走廊场景分类器的硬件实现方案,仿真及实验结果证明了所提基于FPGA的移动机器人SNNs走廊场景分类器的有效性。所提走廊场景分类器不受光照条件的影响,需要的传感器测量信息少,FPGA硬件资源占有率低(LE的利用率仅10%),分类速度快、准确率高,适于实际应用。该研究不仅可以提高移动机器人的环境感知能力和自主性,而且为硬件实现SNNs提供了有益参考。

The neuromorphic-chip is one of the important research aspects of brain-inspired computing,and the hardware implementation of neural networks(NNs)is the basis of neuromorphic-chip.Spiking neural networks(SNNs)with biological plausibility,which convey temporal and spatial information by spikes,are suitable to be implemented with hardware,and are also one of the main tools for brain-inspired computing.A novel SNNs based mobile robots’corridor-scene-classifier implemented by FPGA is proposed.The ultrasonic sensor information of the mobile robot is encoded and input into the SNNs corridor scene classifier,and the corridor scene of the robot is judged by the pulse output mode of the FPGA classifier,so as to improve the environment perception ability and autonomy of the robot.The principle of the Approximation-Spiking IAF neuron model and the implementation of the SNNs corridor-scene-classifier based on Approximation-Spiking IAF by FPGA are discussed in detail.The simulation and experimental results validate the effectiveness of the proposed mobile robots’corridor-scene-classifier based on FPGA and SNNs.Besides the fast processing speed,the classification results of the proposed method are accurate and not influenced by lighting conditions,the needed amount of sensor data is small,the FPGA resource-conquer-rate is low(the utilization rate of LE is only 10%),which is suitable for practical application.Moreover,the proposed corridor classifier can also improve mobile robots’ability of environmental perception and autonomy,and provides a valuable input for SNNs implemented by hardware.