郑重声明:原文参见标题,如有侵权,请联系作者,将会撤销发布!
Front. Neurorobot., (2018)
Abstract
生物智能利用冲动或脉冲来处理信息,这使那些能够在现实世界中感知并采取动作的生物格外出色,并且几乎在生活的各个方面都胜过最先进的机器人。为了弥补这一不足,神经科学、电子学和计算机科学领域新兴的硬件技术和软件知识使人们有可能设计出受大脑机制启发,由SNN控制的生物学上现实的机器人。但是,仍然缺少对基于SNN的控制机器人的全面调研。在本文中,我们调研了过去十年中用于控制任务的SNN领域的发展,并特别关注了快速出现的与机器人相关的应用。首先,我们从速度,能源效率和计算能力方面强调基于SNN的机器人技术的主要动力。然后,我们根据不同的学习规则对那些基于SNN的机器人应用进行分类,并将这些学习规则及其相应的机器人应用进行阐述。我们还简要介绍了一些现有平台,这些平台提供了SNN和机器人仿真之间的交互,以进行探索和开发。最后,我们以对基于SNN的机器人控制方面的未来挑战和一些相关的潜在研究主题的预测来结束调研。
Keywords: spiking neural network, brain-inspired robotics, neurorobotics, learning control, survey
1. INTRODUCTION
2. THEORETICAL BACKGROUND
2.1. Biological Background
2.2. From McCulloch-Pitts to Backpropagation
2.3. Spiking Neural Networks
3. PRIMARY MOTIVATION AND FRAMEWORK
3.1. Primary Impetuses
3.1.1. Biological Plausibility
3.1.2. Speed and Energy Efficiency
3.1.3. Computational Capabilities
3.1.4. Information Processing
3.2. Research Directions
4. MODELING OF SPIKING NEURAL NETWORKS
4.1. Neuron Models
4.2. Information Encoding and Decoding
4.3. Synaptic Plasticity Models
4.3.1. Rate-Based
4.3.2. Spike-Based
4.4. Network Models
4.4.1. Feed-Forward Networks
4.4.2. Recurrent Networks
5. LEARNING AND ROBOTICS APPLICATIONS
5.1. Hebbian-Based Learning
5.1.1. Unsupervised Learning
5.1.2. Supervised Learning
5.1.3. Classical Conditioning
5.1.4. Operant Conditioning
5.1.5. Reward-Modulated Training
5.2. Reinforcement Learning
5.2.1. Temporal Difference
5.2.2. Model-Based
5.3. Others
5.3.1. Evolutionary Algorithms
5.3.2. Self-Organizing Algorithms
5.3.3. Liquid State Machine
6. SIMULATORS AND PLATFORMS
7. OPEN RESEARCH TOPICS
7.1. Biological Mechanism
7.2. Designing and Training SNNs
7.3. High Performance Computing With Neuromorphic Devices
7.4. Interdisciplinary Research of Neuroscience and Robotics
8. CONCLUSION
A Survey of Robotics Control Based on Learning-Inspired Spiking Neural Networks