To ensure the operational integrity of military ground robots in hostile environments, we present a cutting-edge cybersecurity framework that leverages deep learning via a convolutional neural network (CNN). While primarily designed to enhance the security of military ground vehicles, the framework also improves the cybersecurity of the Robot Operating System (ROS), a widely adopted middleware across civilian and military robotic platforms. The GVR-BOT, developed by the United States Army Combat Capabilities Development Command (CCDC) Ground Vehicle Systems Center, serves as the experimental platform. To evaluate the robustness of our system, we conduct real-time penetration testing to identify vulnerabilities within ROS onboard the ground vehicle. Our approach involves training the model on two datasets of normalized network traffic, represented as RGB or grayscale images, to capture vehicle behavior under adversarial conditions. The theoretical foundation is supported by a formal convergence lemma, which guarantees that the CNN model minimizes classification loss under standard smoothness and learning rate assumptions. Experimental results demonstrate the efficacy of our framework, achieving over 98% detection accuracy and under 1% false positive rate, while maintaining minimal detection latency. We also compare different windowing techniques to assess performance trade-offs.

The ability to recognize motions is an important feature in cutting-edge robotics or autonomous systems, such as self-driving cars, humanoid robots, and human-robot interactions, resulting in improved safety and efficiency. Addressing this critical issue, we introduce a simple framework leveraging the benefits of the normalized real-time network traffic data of the middleware ROS platform formulated in the form of RGB or grayscale images to train the Convolutional Neural Network (CNN) system in order to learn the motion pattern of the robot. For our experimental platform, we employ the GVR-BOT Unmanned Ground Vehicle (UGV), developed by the U.S. Army Combat Capabilities Development Command (CCDC), Ground Vehicle Systems Center (GVSC). We rigorously study the performance of our motion recognition system under several different lengths of data (epochs). In addition, we compare the relative merits of our proposed system with respect to the performance of the well-known ‘Bag-of-Features’ (BoFs) detection algorithm widely implemented in computer vision. Our research indicates the efficacy of the proposed motion recognition system as we can achieve a reasonably high detection accuracy ≥ 0.97 within a minimum detection time of two epochs highlighting its real-time benefits. Overall, our recognition system can also achieve superior detection performance compared to the efficacy of the BoFs algorithm.

Safe and secure operations of robotic systems are of paramount importance. Aiming for achieving the trusted operation of a military robotic vehicle under contested environments, we introduce a new cyber-physical system based on the concepts of deep learning convolutional neural networks (CNNs). The proposed algorithm is specifically designed to reduce the cyber vulnerability of the Robot Operating System (ROS), a well-known middleware platform widely used in both civilian and military robots. To demonstrate the efficacy of the proposed algorithm, we conduct penetration testing (real-time man-in-the-middle cyber attack) on the GVR-BOT ground vehicle, a military ground robot, developed by the United States Army Combat Capabilities Development Command (CCDC), Ground Vehicle Systems Center. The cyber attack also exploits the vulnerability of the Robot Operating System (ROS) employed in its onboard computer. We collect experimental data and train our CNN based on two different operating conditions, namely, legitimate and malicious conditions. We normalize and convert the network traffic data in the form of RGB or grayscale images. We introduce two different types of windowing techniques, namely, the independent and overlapping sliding epochs to efficiently feed the network traffic data to our CNN system. Our research indicates the efficacy of the proposed algorithm as our proposed cyber intrusion detection system can achieve reasonably high accuracy of$\geq 99$% and substantially small false-positive rates$\leq$2 % supported with minimum detection time. In addition, we also compare and demonstrate the relative merits of our proposed algorithm with respect to the performance of some well-known techniques, namely, ‘bag-of-features’ and Support Vector Machine (SVM) algorithms.

Recent developments in cutting-edge robotics have been constantly faced with increased cyber-threats, not only in terms of the quantity or the frequency of attacks, but also when it comes to the quality and the severity of the intrusions. This paper provides a systematic overview and critical assessment of state-of-the-art scientific developments in the security aspects of robotics, autonomous systems, and critical infrastructures. Our review highlights open research questions addressing significant research gaps and/or new conceptual frameworks, considering recent advancements in artificial intelligence (AI) and machine learning. Thus the contributions of this paper can be summarised as follows. We first compare and contrast the benefits of multiple cutting-edge AI-based learning algorithms (e.g., fuzzy logic and neural networks) relative to traditional model-based systems (e.g. distributed control and filtering). Subsequently, we point out some specific benefits of AI algorithms to quickly learn and adapt the dynamics of non-linear systems in the absence of complex mathematical models. We also present some potential future research directions (open challenges) in the field. Lastly, this review also delivers an open message to encourage collaborations among experts from multiple disciplines. The implementation of multiple AI algorithms to tackle current security issues in robotics will transform and create novel, hybrid knowledge for intelligent cybersecurity at the application level.

Leveraging the benefits of deep-learning convolutional neural networks, we introduce a new data-driven cyber–physical system specifically designed to address the vulnerability of middleware software, namely, Robot Operating System (ROS), widely implemented in robotics in both civilian and military domains. As for our research platform, we employ the GVR-BOT unmanned ground vehicle, which is a replicate of the U.S. Army ground robot. We focus our study on the ability of the ground robot to operate under fault-tolerant conditions, making it challenging from the standpoint of cybersecurity to differentiate between legitimate and malicious operations. The GVR-BOT ground vehicle belongs to a class of differential drive ground robots and employs ROS in its onboard computer to interact with users. To facilitate deep learning, we develop a large database of images based on the network-traffic data of ROS, describing the dynamics of the GVR-BOT ground robot under legitimate and malicious operations. We use our image database to train and validate the performance of our deep-learning CNN system. Given a set of RGB/grayscale images describing the normalized time-series data representing the dynamics of the GVR-BOT ground robot, the objective of our proposed cybersecurity algorithm is to safeguard the legitimate operation of the ground robot under fault-tolerant conditions, such that any attempts to compromise its performance (e.g., malicious attacks) can be prevented within the minimum detection time. Our research indicates a promising result as our system is capable of detecting malicious attacks with high accuracy while recognizing its legitimate operations with reasonably small false-positive rates.

Moving targets at a very large distance from a camera appear small and of low contrast. The low signal-to-noise-ratio and the presence of clutter in the background degrade the detection performance of conventional moving object detection techniques. To address these challenges, we propose temporal pre-processing of video frames using a biologically-inspired vision model. The bio-inspired model consists of multiple layers of processing analogous to the photoreceptor cells in the visual system of small insects. The adaptive filtering mechanism in the photoreceptor cells suppresses clutter and expands the possible range of input signal changes which improves the target background contrast. We perform experiments on real world video sequences of small moving targets captured with a high bit depth, high resolution and high frame-rate camera. Experimental results show that the biological vision based pre-processing leads to improved detection performance when used in conjunction with a variety of computer vision based moving object detection algorithms. The temporal bio-processing alone has improved the area under the receiver operating characteristic (AUROC) curve of the best performing algorithm by 75.4%. Our results suggest that the bio-inspired pre-processing has strong potential to become a key component of a practical small target detection system.

Computer vision algorithms that make use of optical flow are constantly increasing in complexity, especially in the context of elaborated algorithms that are heavily inspired by biology. To develop upon and utilise these algorithms for realworld tasks, their extensive parameter sets need to be tuned. Due to algorithmic complexity, and non-linearities present throughout their parameter set, this is no small task. Using an adaptive genetic algorithm, which itself is biologically-inspired, we look at the performance and behaviour of the tuning when significant changes have been made to a low speed rotational velocity optical flow estimation algorithm. We validate that previously reported changes to the optical flow estimator yielded a fitness increase of over 30% when compared to the baseline model the changes were made to, and a 15% increase once that baseline model was also tuned. This improvement would be extremely unlikely without the aid of evolutionary computation algorithms. This shows that even an extremely complex computer vision algorithm with many parameters, 36 in this case, can be tuned to an operating point, facilitating the continued development work on the vision algorithm by allowing for the validation and quantification of algorithmic changes.