Zeng Yi's team of the Chinese Academy of Sciences empowers the self-organized obstacle avoidance ability of UAV clusters based on the pulsed neural network of reward regulation Cell Press paper express

Interdisciplinary

Interdisciplinary

On October 28, 2022, the research group of researcher Zeng Yi of the Institute of Automation of the Chinese Academy of Sciences published an article entitled "Nature-inspired self-organizing collision avoidance for drone swarm based on reward-modulated spiking neural" in the Cell Press journal Patterns Network"
.
Inspired by the distributed and self-organizing group intelligence behavior mechanism of biological clusters in nature, they used the reward regulation pulse neural network to realize the online learning of individuals, and independent drone individuals emerged group autonomous obstacle avoidance ability
in the process of self-organizing interaction.

Long press the image recognition QR code to read the original article

Research highlights

• Distributed cluster intelligent behavior
  emerges through local autonomous learning.
  

• UAV individuals independently use the reward-regulated brain-like pulse neural network for online obstacle avoidance learning
  .
  

• UAV swarms demonstrate self-organizing, stable and safe flight
  in a limited area.
  

Introduction to the paper

Swarm behavior is widespread in nature, bees cooperate to find a good source of nectar through swing dance, birds and fish groups spontaneously appear in orderly behavior patterns without collision, and through interaction and cooperation to achieve better predation or resistance to predators
.
The cluster behavior in nature shows the characteristics
of self-organization, decentralization, and distribution.
Each individual independently has a relatively simple ability to learn and interact
with the local environment around them.
The intelligent behavior of the group is derived from the self-organization and coordination between individuals
.
Considering the coupling influence between individual behaviors in computational modeling, the optimization of cluster behavior usually adopts the central control method, while global optimization will bring a lot of computation and poor
adaptability to environmental changes.

The brain-like cognitive intelligence research group led by Zeng Yi, a researcher at the Institute of Automation, Chinese Academy of Sciences, draws on the decentralized and self-organizing behavior mechanism of group intelligence in nature to propose a self-organizing survival obstacle avoidance model
for drone swarms.
Each individual in the cluster independently uses a brain-like pulse neural network to achieve online reinforcement learning, which combines long-term dopamine global regulation and local pulse timing-dependent synaptic plasticity
.
Each individual optimizes the brain-like pulse neural network according to the behavior of other agents observed within a certain range of visual field to achieve efficient and self-organizing interactive learning
.
The intelligent behavior of clusters emerges
self-organizing through local interactions between individuals with online learning capabilities.

Figure 1 Self-organizing obstacle avoidance process of UAV swarm

This model is applied to the survival localization experiment of aphid-like flies, that is, clusters with "territorial ownership" mechanisms can maintain a safe homeostasis with each other in a limited area, without collision and without violating each other's "territory"
.
The results of the survival and locality experiments of different cluster sizes in the simulation scenario show that the model can quickly learn the safe flight strategy and ensure the stable and safe flight
of the cluster for a long time.
The experiment of multiple drones in a limited area in a real scenario also verifies the model's rapid learning and adaptability to dynamic and uncertain environments, and drones can quickly evade each other without collision, as shown
in Figure 2.
Compared with the learning method based on artificial neural network, the model shows better performance and better stability by using pulsed neural network, as shown
in Figure 3.

Figure 2 Demonstration of UAV swarm survival and localization experiment in real scenarios

Fig.
3 Comparison
of results of different methods when the collision threshold is small.
a.
Collisions of
different cluster sizes.
b.
Changes in
the number of collisions between different models during the learning process.

"This study is inspired by the self-organizing and distributed intelligent behavior mechanism of biological clusters in nature, and uses brain-like pulse neural networks with biological rationality combined with local interaction to realize online self-organizing intelligent decision-making
of drone swarms.
" Associate researcher Feifei Zhao, the first author of the paper, said, "From the group behavior decision-making mechanism to the individual online learning model, it is closer to the information processing mechanism of organisms, laying a foundation
for the future development of cluster intelligence that conforms to the learning, decision-making and evolution mechanisms of organisms in nature.
" ”

Researcher Zeng Yi, corresponding author of the paper, said, "We believe that the biggest feature of this study is based on the local brain-like learning and decision-making principles and interaction with the environment, and the evolution and emergence of group-level self-organized obstacle avoidance and safe and stable exploration behavior, which shows that the scientific principles of complex cognitive functions and intelligent behaviors are not necessarily complex, which increases our confidence and determination to
further challenge more complex and higher cognitive functions.
" 。 For nearly a decade, we have been continuously building a brain-inspired cognitive intelligence engine (BrainCog) for fully pulsed neural networks to support the decoding of the essence of biological intelligence, including humans, and to develop brain-like artificial intelligence
on this basis 。 The research in this paper is the basic exploration and application of the brain-like cognitive intelligence engine BrainCog in the brain-like learning mechanism and emergence, behavior evolution, and we open source all the relevant models and algorithms, hoping to promote the collaborative development of brain-like artificial intelligence with the academic community.
"

About the author

Feifei Zhao

Associate Researcher

Feifei Zhao is an associate researcher
in the brain-like cognitive intelligence research group of the Institute of Automation, Chinese Academy of Sciences.
His research interests include brain-like decision-making, development and evolution of pulsed neural networks
.
He has published many papers in Patterns, IEEE Transactions on Cognitive and Developmental Systems, Neural Computation, Scientific Reports, Cognitive Computation, Frontiers in Neuroscience
, etc.

Zeng Yi

researcher

Zeng Yi, researcher of the Institute of Automation, Chinese Academy of Sciences, head of the brain-like cognitive intelligence research group, deputy director of the Brain Atlas and Brain-like Intelligence Laboratory, director of the Artificial Intelligence Ethics and Governance Research Center; Professor, doctoral supervisor, University of Chinese Academy of Sciences; Member of the National New Generation Artificial Intelligence Governance Committee; Expert of UNESCO's Ad Hoc Expert Group on the Ethics of
Artificial Intelligence.
His research interests include: brain-like artificial intelligence, artificial intelligence ethics, governance and sustainable development
.
Representative results have been published in Cell Press journals Patterns, iScience and Scientific Reports, Science Advances, IEEE Transactions, and IJCAI and AAAI
, an important international conference in the field of artificial intelligence.

Han Bing

PhD candidate

Han Bing, 2021 PhD candidate of the Brain-like Cognitive Intelligence Research Group, Institute of Automation, Chinese Academy of Sciences, under the supervision of researcher
Zeng Yi.
His research interests are brain-like developmental pulsed neural networks
.
He has published a paper
in Patterns.

Fang Hongjian

doctor

Fang Hongjian, 2017 doctoral candidate of the Brain-like Cognitive Intelligence Research Group, Institute of Automation, Chinese Academy of Sciences, under the supervision of researcher
Zeng Yi.
His research interests include brain-like pulse neural networks, symbolic representation, causal reasoning, etc
.
He has published many papers in Patterns, Frontiers in Computational Neuroscience, IJCNN
, etc.

Zhao Zhuoya

PhD candidate

Zhao Zhuoya is a 2019 doctoral candidate in the Brain-like Cognitive Intelligence Research Group of the Institute of Automation, Chinese Academy of Sciences, under the supervision of researcher
Zeng Yi.
His research interests are brain-like thinking speculation and decision-making models
.
He has published many papers in Patterns, Frontiers in Neuroscience
, etc.

Related paper information

The research results are published in Cell Press's journal Patterns, click "Read More" or scan the QR code below to view the paper
.

▌Papers:

Nature-inspired Self-organizing Collision Avoidance for Drone Swarm Based on Reward-modulated Spiking Neural Network

▌Paper URL:

style="white-space: normal;box-sizing: border-box;" _msthash="251170" _msttexthash="10239255">▌ DOI：

https://doi.
org/10.
1016/j.
patter.
2022.
100611

Long press the image recognition QR code to read the original article

In 1974, we published our first flagship journal, Cell
.
Today, CellPress has developed into an international cutting-edge academic publishing
house in the field of science with more than 50 journals.
We firmly believe that the power of science will always benefit mankind
.

CellPress Cell Press