Spiking neural network-based navigation and obstacle avoidance algorithm for complex scenes

doi:10.11996/JG.j.2095-302X.2023061121

Abstract

Abstract:

Spiking neural network (SNN) have been widely applied in the field of mobile robot navigation and obstacle avoidance due to their low power consumption and temporal processing capabilities. However, existing SNN models are relatively simple and struggle with addressing obstacle avoidance in complex scenarios, such as dynamic obstacles with varying speeds and environmental noise interference. To tackle these challenges, a complex scene navigation and obstacle avoidance algorithm was proposed based on SNNs. This algorithm employed attention mechanisms to enhance obstacle avoidance capabilities for dynamic obstacles, enabling the model to make more accurate obstacle avoidance decisions by focusing more on the information of dynamic obstacles. Additionally, a dynamic spiking threshold was designed based on biological inspiration, allowing the model to adaptively adjust the firing of spiking signals to adapt to environments with noise interference. Experimental results demonstrated that the proposed algorithm exhibited optimal navigation and obstacle avoidance performance within virtual complex scenes. Across the three designed complex scenes (variable-speed dynamic scenes, input interference, and weight interference), the navigation obstacle avoidance success rates could reach 86.5%, 79.0%, and 76.2%, respectively. This research provided a new approach and method for solving the problem of robot navigation and obstacle avoidance in complex scenarios.

Key words: spiking neural network, navigation and obstacle avoidance, mobile robot, dynamic spiking threshold, attention mechanism

CLC Number:

TP391

DING Jian-chuan, XIAO Jin-tong, ZHAO Ke-xin, JIA Dong-qing, CUI Bing-de, YANG Xin. Spiking neural network-based navigation and obstacle avoidance algorithm for complex scenes[J]. Journal of Graphics, 2023, 44(6): 1121-1129.

Figures/Tables 10

Fig. 1 Complex navigation and obstacle avoidance scenarios

Fig. 2 Fitting of pseudo-gradient functions ((a) Impulse function; (b) Pseudo-gradient function)

Fig. 3 Algorithm flow for complex scene navigation and obstacle avoidance

Fig. 4 Structure of the spiking attention module

Fig. 5 Spiking firing threshold ((a) Static spiking threshold; (b) Dynamic spiking threshold)

Fig. 6 Dynamic spiking firing threshold model

Fig. 7 Gazebo scenes ((a) Training scene; (b) Testing scene)

Table 1 Navigation obstacle avoidance success rate of different models in complex scenarios (%)

模型	变速动态场景	输入干扰	权重干扰
ROS^[21]	62.0	51.5	-
SAN^[9]	79.5	70.0	52.6
BDETT^[16]	84.0	78.4	77.0
Ours	86.5	79.0	76.2

Fig. 8 Heatmap of obstacle avoidance test

Table 2 The ablation experiment of our algorithm in complex scenarios (%)

模型			变速动态场景	输入干扰	权重干扰
SAN^[9]	SAM	DTH	变速动态场景	输入干扰	权重干扰
√	-	-	79.5	70.0	51.6
√	√	-	85.5	72.5	51.3
√	-	√	83.0	79.0	76.0
√	√	√	86.5	79.0	76.2

References 22

[1]	张巧荣, 崔明义. 基于改进Dijkstra算法的机器人路径规划方法[J]. 微计算机信息, 2007, 23(2): 286-287, 136.
	ZHANG Q R, CUI M Y. A dijkstra algorithm based approach for robot path planning[J]. Microcomputer Information, 2007, 23(2): 286-287, 136 (in Chinese).
[2]	周宇杭, 王文明, 李泽彬, 等. 基于A星算法的移动机器人路径规划应用研究[J]. 电脑知识与技术, 2020, 16(13): 1-3, 10.
	ZHOU Y H, WANG W M, LI Z B, et al. Application research of mobile robot path planning based on A-star algorithm[J]. Computer Knowledge and Technology, 2020, 16(13): 1-3, 10 (in Chinese).
[3]	FOX D, BURGARD W, THRUN S. The dynamic window approach to collision avoidance[J]. IEEE Robotics & Automation Magazine, 1997, 4(1): 23-33.
[4]	TAI L, LI S H, LIU M. A deep-network solution towards model-less obstacle avoidance[C]// 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems. New York: IEEE Press, 2016: 2759-2764.
[5]	PFEIFFER M, SHUKLA S, TURCHETTA M, et al. Reinforced imitation: sample efficient deep reinforcement learning for mapless navigation by leveraging prior demonstrations[J]. IEEE Robotics and Automation Letters, 2018, 3(4): 4423-4430. DOI URL
[6]	LONG P X, FAN T X, LIAO X Y, et al. Towards optimally decentralized multi-robot collision avoidance via deep reinforcement learning[C]// 2018 IEEE International Conference on Robotics and Automation. New York: IEEE Press, 2018: 6252-6259.
[7]	CHOI J, PARK K, KIM M, et al. Deep reinforcement learning of navigation in a complex and crowded environment with a limited field of view[C]// 2019 International Conference on Robotics and Automation. New York: IEEE Press, 2019: 5993-6000.
[8]	MILDE M B, BLUM H, DIETMÜLLER A, et al. Obstacle avoidance and target acquisition for robot navigation using a mixed signal analog/digital neuromorphic processing system[J]. Frontiers in Neurorobotics, 2017, 11: 28. DOI PMID
[9]	TANG G Z, KUMAR N, MICHMIZOS K P. Reinforcement co-learning of deep and spiking neural networks for energy-efficient mapless navigation with neuromorphic hardware[C]// 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems. New York: IEEE Press, 2020: 6090-6097.
[10]	WANG X Q, HOU Z G, LV F, et al. Mobile robots' modular navigation controller using spiking neural networks[J]. Neurocomputing, 2014, 134: 230-238. DOI URL
[11]	张军军. 基于注意力机制卷积脉冲神经网络的目标识别方法[J]. 计算机与数字工程, 2022, 50(9): 1956-1961.
	ZHANG J J. Object recognition method based on attention mechanism convolutional spiking neural networks[J]. Computer & Digital Engineering, 2022, 50(9): 1956-1961 (in Chinese).
[12]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[13]	WANG Q L, WU B G, ZHU P F, et al. ECA-net: efficient channel attention for deep convolutional neural networks[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 11534-11542.
[14]	BEAR M F. Mechanism for a sliding synaptic modification threshold[J]. Neuron, 1995, 15(1): 1-4. PMID
[15]	ZHANG W, LINDEN D J. The other side of the engram: experience-driven changes in neuronal intrinsic excitability[J]. Nature Reviews Neuroscience, 2003, 4(11): 885-900. DOI PMID
[16]	DING J C, DONG B, HEIDE F, et al. Biologically inspired dynamic thresholds for spiking neural networks[EB/OL]. [2022-12-20]. https://www.zhuanzhi.ai/paper/75c640a8c462ef30d06c2ff1e6525f35.
[17]	WU Y J, DENG L, LI G Q, et al. Spatio-temporal backpropagation for training high-performance spiking neural networks[J]. Frontiers in Neuroscience, 2018, 12: 331-342. DOI PMID
[18]	DESAI N S, RUTHERFORD L C, TURRIGIANO G G. Plasticity in the intrinsic excitability of cortical pyramidal neurons[J]. Nature Neuroscience, 1999, 2(6): 515-520. PMID
[19]	HIGGS M H, SPAIN W J. Kv1 channels control spike threshold dynamics and spike timing in cortical pyramidal neurones[J]. The Journal of Physiology, 2011, 589(Pt 21): 5125-5142. DOI PMID
[20]	FLORENSA C, HELD D, WULFMEIER M, et al. Reverse curriculum generation for reinforcement learning[EB/OL]. [2022-12-20]. https://arxiv.org/abs/1707.05300.
[21]	GUIMARÃES R L, DE OLIVEIRA A S, FABRO J A, et al. ROS navigation: concepts and tutorial[M]// Robot Operating System (ROS). Cham: Springer, 2016: 121-160.
[22]	PAGKALOS M, CHAVLIS S, POIRAZI P. Introducing the Dendrify framework for incorporating dendrites to spiking neural networks[J]. Nature Communications, 2023, 14(1): 131. DOI PMID