Bio-inspired Neural Network Research Articles

Dynamic path planning of autonomous bulldozers using activity-value-optimised bio-inspired neural networks and adaptive cell decomposition

Autonomous bulldozers encounter critical challenges in dynamic path planning in complex construction environments such as dynamic obstacles and narrow passages. Bio-inspired neural networks (BINN) are commonly employed for real-time path planning in dynamic environments. However, the algorithm is not feasible enough to calculate the positive input of neighbour cells, thus resulting in unreasonable distribution of activity values in the obstacle-dense regions. Additionally, the use of regular grids results in excessive computation. These drawbacks cause the algorithm to plan irrational paths and reduce computational efficiency. This study proposed an activity-value-optimised BINN and adaptive cell decomposition for the dynamic path planning of autonomous bulldozers. First, an adaptive cell decomposition method was proposed to model a highly intricate and constantly evolving environment, resulting in a reduction in the computational workload of the BINN. Second, the calculation of the activity value is optimised by enhancing the shunt equation of the BINN, which considers both obstacles and grid levels. This improvement enabled more reasonable distribution results. The smoothness of dynamic path planning is refined by incorporating the turning radius, which can be more effectively integrated with local path planning. Compared with the original BINN algorithm and other baseline algorithms, a practical large-scale hydropower project application demonstrated that the proposed method can effectively reduce the number of computational iterations, enrich the travel direction, and improve path quality while ensuring dynamic obstacle avoidance.

MEMS reservoir computing system with stiffness modulation for multi-scene data processing at the edge

Reservoir computing (RC) is a bio-inspired neural network structure which can be implemented in hardware with ease. It has been applied across various fields such as memristors, and electrochemical reactions, among which the micro-electro-mechanical systems (MEMS) is supposed to be the closest to sensing and computing integration. While previous MEMS RCs have demonstrated their potential as reservoirs, the amplitude modulation mode was found to be inadequate for computing directly upon sensing. To achieve this objective, this paper introduces a novel MEMS reservoir computing system based on stiffness modulation, where natural signals directly influence the system stiffness as input. Under this innovative concept, information can be processed locally without the need for advanced data collection and pre-processing. We present an integrated RC system characterized by small volume and low power consumption, eliminating complicated setups in traditional MEMS RC for data discretization and transduction. Both simulation and experiment were conducted on our accelerometer. We performed nonlinearity tuning for the resonator and optimized the post-processing algorithm by introducing a digital mask operator. Consequently, our MEMS RC is capable of both classification and forecasting, surpassing the capabilities of our previous non-delay-based architecture. Our method successfully processed word classification, with a 99.8% accuracy, and chaos forecasting, with a 0.0305 normalized mean square error (NMSE), demonstrating its adaptability for multi-scene data processing. This work is essential as it presents a novel MEMS RC with stiffness modulation, offering a simplified, efficient approach to integrate sensing and computing. Our approach has initiated edge computing, enabling emergent applications in MEMS for local computations.

Complete area-coverage path planner for surface cleaning in hospital settings using mobile dual-arm robot and GBNN with heuristics

Complete area-coverage path planners are essential for robots performing tasks like cleaning, inspection, and surveying. However, they often involve complex calculations, mapping, and determining movement directions, leading to high computational or processing overheads and the risk of deadlocks. This paper proposes an approach for cleaning, i.e., by linear wiping of generic and discontinuous surfaces in hospital settings using inhouse assembled mobile dual-arm (MDA) robotic system. The proposed framework introduces key features: (a) a less resource-intensive approach for MDA positioning and cleaning surface mapping, (b) Modified Glasius Bioinspired Neural Network through use of heuristics (GBNN+H) to optimize surface linear wiping while obstacle avoidance, and traversal across discontinuous surfaces. The advantages of the proposed algorithm are highlighted in simulation with GBNN+H significantly reduces the number of steps and flight time required for complete coverage compared to existing algorithms. The proposed framework is also experimentally demonstrated in a hospital setting, paving the way for improved automation in cleaning and disinfection tasks. Overall, this work presents a generic and versatile, applicable to various surface orientations and complexities.

IMUPF-BIN: A new autonomous search method for radioactive sources

To solve the problem of mobile robots autonomously searching for unknown radioactive sources, this paper proposes an automatic source search strategy, abbreviated as IMUPF-BIN: the location of the radioactive source is estimated by improved unscented particle filtering (IMUPF) algorithm, and the search path is predicted and planned by bioinspired neural network (BIN) model. First, the search environment is expressed as an occupied grid map, and the BIN model is combined with the grid occupied map to obtain the environment information in the search process. Then, the unscented particle filter (UPF) which is used for source parameter estimation is improved by introducing the idea of linear optimization to improve the efficiency of particle usage and the diversity of particles. Further, the BIN model is introduced to reduce the influence of obstacles on the movement of the robot, thereby improving the search accuracy and efficiency of radioactive source. After the robot detects the radiation count value, the parameter information of the source is estimated through the IMUPF algorithm. By using the activity value of the BIN model and distributed model predictive control (DMPC) to plan the path, while continuously iterating the search process, the location of the radioactive source is finally determined. The simulation results show that in without obstacles environment, the localization accuracy of the IMUPF algorithm is improved by 85.6% and 52.8% compared to the standard particle filtering (PF) and UPF algorithm respectively, and the search time of the IMUPF-BIN algorithm is reduced by 75.9% compared to the gradient search (GS) algorithm. In obstacles environment, the localization accuracy of the IMUPF algorithm is improved by 80.8% and 35.9% compared to the PF and UPF algorithms respectively. In addition, the GS algorithm fails to search for the radioactive source due to the shield of obstacles, while the IMUPF-BIN algorithm can successfully search for radioactive source in a shorter search time.

Multi-AUV underwater static target search method based on consensus-based bundle algorithm and improved Glasius bio-inspired neural network

This research introduces a hierarchical strategy for static target searches with multi-autonomous underwater vehicles (AUVs) to optimize cumulative search rewards. The approach comprises two primary elements: task allocation and path planning. A Voronoi diagram segments regions based on peak detection via a maximum filter in the task allocation stage. Then, a consensus-based bundling algorithm ensures the load-balanced distribution of peak sub-regions across AUVs, while a dynamic cooperation mechanism allows for dynamic adjustment of task allocation, thereby increasing the system's operational flexibility. Path planning employs an improved Glasius bio-inspired neural network, leveraging analogies to convolution processes and incorporating mean pooling, multiple convolutions, and resampling. This method enhances global information propagation and optimizes path point selection through a discounted reward function evaluating adjacent nodes, thus boosting the search efficiency of individual AUVs. Simulation experiments validate the method's effectiveness and robustness in multi-AUV static target searches, demonstrating its potential to improve search efficiency.

Surrogate gradient learning in spiking networks trained on event-based cytometry dataset

Spiking neural networks (SNNs) are bio-inspired neural networks that - to an extent - mimic the workings of our brains. In a similar fashion, event-based vision sensors try to replicate a biological eye as closely as possible. In this work, we integrate both technologies for the purpose of classifying micro-particles in the context of label-free flow cytometry. We follow up on our previous work in which we used simple logistic regression with binary labels. Although this model was able to achieve an accuracy of over 98%, our goal is to utilize the system for a wider variety of cells, some of which may have less noticeable morphological variations. Therefore, a more advanced machine learning model like the SNNs discussed here would be required. This comes with the challenge of training such networks, since they typically suffer from vanishing gradients. We effectively apply the surrogate gradient method to overcome this issue achieving over 99% classification accuracy on test data for a four-class problem. Finally, rather than treating the neural network as a black box, we explore the dynamics inside the network and make use of that to enhance its accuracy and sparsity.

A sensor system integrating sensing and intelligence based on MEMS reservoir computing

Reservoir computing (RC) is a bio-inspired neural network structure which is easy to be implemented in hardware. It has been constructed in a great many fields such as memristor, electrochemical reaction, among which MEMS is the closest to integrate sensing and computing. We propose a novel sensor system of MEMS RC based on stiffness modulation, that natural signal directly affects the system stiffness as the input. Under this paradigm, information can be processed locally without data collection and pre-processing. We inherited the nonlinearity tuning principle and optimized the post-processing algorithm by creating a digital mask operator. In this way, the system can deal with classification tasks as well as forecasting tasks. We integrated MEMS, IC and FPGA with a small volume and low power consumption, so complicated setup for data discretization and transduction in traditional MEMS RC is eliminated. The system can process word classification and chaos forecasting with high accuracy, which preliminarily proves the integrated RC architecture.

Inter-Reconfigurable Robot Path Planner for Double-Pass Complete Coverage Problem

Recent advancements in autonomous mobile robots have led to significant progress in area coverage tasks. However, challenges persist in optimizing the efficiency and computational complexity of complete coverage path planner (CCPP) algorithms for multi-robot systems, particularly in scenarios requiring revisiting or a double pass in specific locations, such as cleaning robots addressing spilled consumables. This paper presents an innovative approach to tackling the double-pass complete coverage problem using an autonomous inter-reconfigurable robot path planner. Our solution leverages a modified Glasius bio-inspired neural network (GBNN) to facilitate double-pass coverage through inter-reconfiguration between two robots. We compare our proposed algorithm with traditional multi-robot path planning in a centralized system, demonstrating a reduction in algorithm iterations and computation time. Our experimental results underscore the efficacy of the proposed solution in enhancing the efficiency of area coverage tasks. Furthermore, we discuss the implementation details and limitations of our study, providing insights for future research directions in autonomous robotics.

Bio-Inspired Neural Network for Real-Time Evasion of Multi-Robot Systems in Dynamic Environments.

In complex and dynamic environments, traditional pursuit-evasion studies may face challenges in offering effective solutions to sudden environmental changes. In this paper, a bio-inspired neural network (BINN) is proposed that approximates a pursuit-evasion game from a neurodynamic perspective instead of formulating the problem as a differential game. The BINN is topologically organized to represent the environment with only local connections. The dynamics of neural activity, characterized by the neurodynamic shunting model, enable the generation of real-time evasive trajectories with moving or sudden-change obstacles. Several simulation and experimental results indicate that the proposed approach is effective and efficient in complex and dynamic environments.

Credit and Loan Approval Classification Using a Bio-Inspired Neural Network.

Numerous people are applying for bank loans as a result of the banking industry's expansion, but because banks only have a certain amount of assets to lend to, they can only do so to a certain number of applicants. Therefore, the banking industry is very interested in finding ways to reduce the risk factor involved in choosing the safe applicant in order to save lots of bank resources. These days, machine learning greatly reduces the amount of work needed to choose the safe applicant. Taking this into account, a novel weights and structure determination (WASD) neural network has been built to meet the aforementioned two challenges of credit approval and loan approval, as well as to handle the unique characteristics of each. Motivated by the observation that WASD neural networks outperform conventional back-propagation neural networks in terms of sluggish training speed and being stuck in local minima, we created a bio-inspired WASD algorithm for binary classification problems (BWASD) for best adapting to the credit or loan approval model by utilizing the metaheuristic beetle antennae search (BAS) algorithm to improve the learning procedure of the WASD algorithm. Theoretical and experimental study demonstrate superior performance and problem adaptability. Furthermore, we provide a complete MATLAB package to support our experiments together with full implementation and extensive installation instructions.

Intelligent Fish-Inspired Foraging of Swarm Robots with Sub-Group Behaviors Based on Neurodynamic Models.

This paper proposes a novel intelligent approach to swarm robotics, drawing inspiration from the collective foraging behavior exhibited by fish schools. A bio-inspired neural network (BINN) and a self-organizing map (SOM) algorithm are used to enable the swarm to emulate fish-like behaviors such as collision-free navigation and dynamic sub-group formation. The swarm robots are designed to adaptively reconfigure their movements in response to environmental changes, mimicking the flexibility and robustness of fish foraging patterns. The simulation results show that the proposed approach demonstrates improved cooperation, efficiency, and adaptability in various scenarios. The proposed approach shows significant strides in the field of swarm robotics by successfully implementing fish-inspired foraging strategies. The integration of neurodynamic models with swarm intelligence not only enhances the autonomous capabilities of individual robots, but also improves the collective efficiency of the swarm robots.

Efficient Coverage Path Planning and Underwater Topographic Mapping of an USV based on A*-Improved Bio-Inspired Neural Network

Efficient Coverage Path Planning and Underwater Topographic Mapping of an USV based on A*-Improved Bio-Inspired Neural Network

A Novel Feature Learning-Based Bio-Inspired Neural Network for Real-Time Collision-Free Rescue of Multirobot Systems

A Novel Feature Learning-Based Bio-Inspired Neural Network for Real-Time Collision-Free Rescue of Multirobot Systems

Bio-inspired affordance learning for 6-DoF robotic grasping: A transformer-based global feature encoding approach

The 6-Degree-of-Freedom (6-DoF) robotic grasping is a fundamental task in robot manipulation, aimed at detecting graspable points and corresponding parameters in a 3D space, i.e affordance learning, and then a robot executes grasp actions with the detected affordances. Existing research works on affordance learning predominantly focus on learning local features directly for each grid in a voxel scene or each point in a point cloud scene, subsequently filtering the most promising candidate for execution. Contrarily, cognitive models of grasping highlight the significance of global descriptors, such as size, shape, and orientation, in grasping. These global descriptors indicate a grasp path closely tied to actions. Inspired by this, we propose a novel bio-inspired neural network that explicitly incorporates global feature encoding. In particular, our method utilizes a Truncated Signed Distance Function (TSDF) as input, and employs the recently proposed Transformer model to encode the global features of a scene directly. With the effective global representation, we then use deconvolution modules to decode multiple local features to generate graspable candidates. In addition, to integrate global and local features, we propose using a skip-connection module to merge lower-layer global features with higher-layer local features. Our approach, when tested on a recently proposed pile and packed grasping dataset for a decluttering task, surpassed state-of-the-art local feature learning methods by approximately 5% in terms of success and declutter rates. We also evaluated its running time and generalization ability, further demonstrating its superiority. We deployed our model on a Franka Panda robot arm, with real-world results aligning well with simulation data. This underscores our approach’s effectiveness for generalization and real-world applications.

A Bio-Inspired Approach to Enhancing Machine Learning: Integrating Spiking Neural Networks

In this study, we investigate the integration of bio-inspired neural networks, specifically spiking neural networks (SNNs), into conventional machine learning frameworks. Known for their energy efficiency and ability to process temporal information, SNNs present unique advantages over traditional artificial neural networks (ANNs). This paper explores recent advancements in SNNs, focusing on methodologies such as spike-timing-dependent plasticity (STDP), hybrid conversion techniques, and learnable membrane time constants [1][2]. We evaluate the application of SNNs in visual categorization, digit recognition, and biomedical imaging [3][4]. A comprehensive literature review identifies current strengths and lim

Advancements in Complementary Metal-Oxide Semiconductor-Compatible Tunnel Barrier Engineered Charge-Trapping Synaptic Transistors for Bio-Inspired Neural Networks in Harsh Environments.

This study aimed to propose a silicon-on-insulator (SOI)-based charge-trapping synaptic transistor with engineered tunnel barriers using high-k dielectrics for artificial synapse electronics capable of operating at high temperatures. The transistor employed sequential electron trapping and de-trapping in the charge storage medium, facilitating gradual modulation of the silicon channel conductance. The engineered tunnel barrier structure (SiO2/Si3N4/SiO2), coupled with the high-k charge-trapping layer of HfO2 and high-k blocking layer of Al2O3, enabled reliable long-term potentiation/depression behaviors within a short gate stimulus time (100 μs), even under elevated temperatures (75 and 125 °C). Conductance variability was determined by the number of gate stimuli reflected in the maximum excitatory postsynaptic current (EPSC) and the residual EPSC ratio. Moreover, we analyzed the Arrhenius relationship between the EPSC as a function of the gate pulse number (N = 1-100) and the measured temperatures (25, 75, and 125 °C), allowing us to deduce the charge trap activation energy. A learning simulation was performed to assess the pattern recognition capabilities of the neuromorphic computing system using the modified National Institute of Standards and Technology datasheets. This study demonstrates high-reliability silicon channel conductance modulation and proposes in-memory computing capabilities for artificial neural networks using SOI-based charge-trapping synaptic transistors.

Improved BINN-Based Underwater Topography Scanning Coverage Path Planning for AUV in Internet of Underwater Things

Deep understanding the special nature of underwater topography plays an important role for Internet of Underwater Things (IoUT). Nowadays, underwater topography scanning with Autonomous Underwater Vehicle (AUV) has been becoming the chief methodology of knowing Seabed topography and geomorphology. How to design topography scanning trajectory can be mathematically described as a full Coverage Path Planning (CPP) problem. In this paper, facing the complete coverage path planning problem of mobile AUV, a new strategy based on Bio-Inspired Neural Network (BINN) algorithm with improved activity value of each neuron is discussed in detail. The original activity value function in BINN is instead of a piecewise linear function to reduce computational complexity. In addition, to overcome traditional dead-zone problem, an A* path planning based dead-zone escape method along the shorter path as early as possible to the recently uncovered area is described in deep. Extensive simulation results and practical experiments verify the performance of proposed Improved Bio-Inspired Neural Network (IBINN in short) based algorithm.

Artificial nanophotonic neuron with internal memory for biologically inspired and reservoir network computing

Neurons with internal memory have been proposed for biological and bio-inspired neural networks, adding important functionality. We introduce an internal time-limited charge-based memory into a III–V nanowire (NW) based optoelectronic neural node circuit designed for handling optical signals in a neural network. The new circuit can receive inhibiting and exciting light signals, store them, perform a non-linear evaluation, and emit a light signal. Using experimental values from the performance of individual III–V NWs we create a realistic computational model of the complete artificial neural node circuit. We then create a flexible neural network simulation that uses these circuits as neuronal nodes and light for communication between the nodes. This model can simulate combinations of nodes with different hardware derived memory properties and variable interconnects. Using the full model, we simulate the hardware implementation for two types of neural networks. First, we show that intentional variations in the memory decay time of the nodes can significantly improve the performance of a reservoir network. Second, we simulate the implementation in an anatomically constrained functioning model of the central complex network of the insect brain and find that it resolves an important functionality of the network even with significant variations in the node performance. Our work demonstrates the advantages of an internal memory in a concrete, nanophotonic neural node. The use of variable memory time constants in neural nodes is a general hardware derived feature and could be used in a broad range of implementations.

A Multirobot Distributed Collaborative Region Coverage Search Algorithm Based on Glasius Bio-Inspired Neural Network

There are many constraints for a multi-robot system to perform a region coverage search task in an unknown environment. To address this, we propose a novel multi-robot distributed collaborative region coverage search algorithm based on Glasius bio-inspired neural network (GBNN). Firstly, we develop an environmental information updating model to represent the dynamic search environment. This model converts the environmental information detected by the robot into dynamic neural activity landscape of GBNN. Secondly, we introduce the distributed model predictive control method in search path planning to improve search efficiency. In addition, we propose a distributed collaborative decision-making mechanism among the robots to produce several dynamic search sub-teams. Within each sub-team, collaborative decisions are made among the robot members to optimize the solution and obtain the next movement path of each robot. Finally, we conduct experiments in three aspects to verify the effectiveness of the proposed method. Compared with three algorithms in this field, the experimental results demonstrate that the proposed algorithm exhibits good performance in a multi-robot region coverage search task.

Path Planning for Multiple Unmanned Surface Vehicles Using Glasius Bio-Inspired Neural Network With Hungarian Algorithm

In this article, we focus on the cooperative path planning for unmanned surface vehicles (USVs) composed of single-USV path planning and multi-USVs task assignment. First, the Glasius bioinspired neural network (GBNN) is used to calculate the neural activity for the discretized working space of USV, and the ocean current is considered in the definition of neuron connection weight especially. The standard path of single USV for obstacle avoidance between start point and destination can hence be planned. Then, based on the result of neural activity values from GBNN, the cost matrix of the Hungarian algorithm is built and modified for the task assignment among multi-USVs, and the unbalanced problem is well resolved. Consequently, the task points are allocated to USVs and prioritized effectively, and each USV just needs to visit the corresponding task points respectively along the standard paths by GBNN to accomplish the task. Finally, the simulation results demonstrate the feasibility and efficiency of the proposed algorithm.