This PDF file contains the front matter associated with SPIE Proceedings Volume 11425, including the Title Page, Copyright information, and Table of Contents.

A tightly integrated image-aided Inertial Navigation System (INS), which copes with GNSS failure, is tested with a realistic data set from an Octocopter. The system integrates the inertial sensor data with position tracks of image feature points over an image sequence in an error-state extended Kalman filter (EKF). The Octocopter is equipped with a rig of three cameras in the horizontal direction with overlapping fields of view. Our main aim is to utilize the data from the three cameras as a single-sensor data. However, as an intermediate experiment, the cameras are considered as three individual sensors and the performance of the image-aided INS with the different combinations of data integration is analyzed in this study. The image-aided INS reduced the drift drastically compared to the drift in free-inertial when integrating the image data sets separately or in combinations. However, the combination of all three data sets together performed poorer than the other combinations, probably due to correlated errors that are not adequately modeled by the current EKF.

The recurrent monitoring of an aerial vehicle for structural damage detection and identification by acoustic sensors increases its reliability and remaining useful lifetime. In order to reach full structural health monitoring (SHM) autonomy, there is a need to combine sensing and communication functions into smart multifunctional sensors. Through this fusion, the information gathered by the SHM sensors could be transmitted in real time to a central processing unit without any human intervention. To that end, this research paper proposes reusing the existing network of acoustic SHM sensors mounted or embedded in the structure to enable acoustic multi-sensor wireless communication through the structure itself using elastic waves as the carrier signals. By doing so, the proposed acoustic communication system does not generate additional radio-frequency (RF) interference to other RF communication systems on board such as those used for vehicle control and safety-related services. This paper describes the design of the proposed acoustic wireless sensor network for autonomous SHM of aerial vehicles. First, the network topology and sensors placement are described along with the data routing algorithm. Then, the time-reversal based time division multiple access technique is introduced for multi-sensor communication using elastic waves. The data transmission across the elastic channel using time-reversal pulse position modulation is also presented. Finally, the system is evaluated based on the acoustic channel response of the horizontal stabilizer of an Ercoupe 415-C aircraft.

This article describes a new approach for distributed 3D SLAM map building. The key contribution of this article is the creation of a distributed graph-SLAM map-building architecture responsive to bandwidth and computational needs of the robotic platform. Responsiveness is afforded by integration of a 3D point cloud to plane cloud compression algorithm that approximates dense 3D point cloud using local planar patches. Compute bound platforms may restrict the computational duration of the compression algorithm and low-bandwidth platforms can restrict the size of the compression result. The backbone of the approach is an ultra-fast adaptive 3D compression algorithm that transforms swaths of 3D planar surface data into planar patches attributed with image textures. Our approach uses DVO, a leading algorithm for 3D mapping, and extends it by computationally isolating map integration tasks from local Guidance, Navigation and Control tasks and includes an addition of a network protocol to share the compressed planes. The joint effect of these contributions allows agents with 3D sensing capabilities to calculate and communicate compressed map information commensurate with their on-board computational resources and communication channel capacities. This opens SLAM mapping to new categories of robotic platforms that may have computational and memory limits that prohibit other SLAM solutions.

In this paper, we propose decentralized position controllers for a team of point-mass robots that must cooperatively transport a payload to a target location. The robots have double-integrator dynamics and are rigidly attached to the payload. The controllers only require robots' measurements of their own positions and velocities, and the only information provided to the robots is the desired position of the payload's center of mass. We consider scenarios in which the robots do not know the position of the payload's center of mass and try to selfishly stabilize their own positions to the desired location, similar to the behaviors exhibited by certain species of ants when retrieving food items in groups. We propose a proportional-derivative (PD) controller that does not rely on inter-robot communication, prior information about the load dynamics and geometry, or knowledge of the number of robots and their distribution around the payload. Using a Lyapunov argument, we prove that under this control strategy, the payload's center of mass converges to a neighborhood of the desired position. Moreover, we prove that the payload's rotation is bounded, and its angular velocity converges to zero. We show that the error between the steady-state position of the payload's center of mass and its desired position depends on the robots' distribution around the payload's center of mass, with a uniform distribution resulting in the lowest steady-state error. We validate our theoretical results with simulations in MATLAB.

In this paper, we report our preliminary simulation-based efforts in designing feedback for the thruster-assisted walking of a bipedal robot, called Harpy, currently being developed at Northeastern University. The biped is equipped with a total of eight actuators, and two pairs of coaxial thrusters fixed to its torso. Each leg is equipped with three actuated joints, the actuators located at the hip allow the legs to move sideways and actuation in the lower portion of the legs is realized through a parallelogram mechanism. Two extra actuators rotate the thrusters with respect to the torso, therefore, they provide more flexibility in control.

Safety in autonomous vehicles is a challenging task because it depends on many factors such as weather conditions, sensors, the complexity of the surrounding environment and many more. These factors are unpredictable and hard to capture in real life. Automated vehicle systems depend on sensors such as LiDAR, radar and cameras to enable them to reach safely to the target destination. In this study, we show how the automated vehicle system that utilizes a radar and a camera as an input to the Pedestrian Protection System (PPS) is influenced by uncertain weather conditions and sensor failure. Under these conditions, we investigate surrogate safety measures such as Pedestrian Classification Time-to-Collision (PCT), and Post encroachment (PET). This study uses a physics-based simulation software called Prescan as well as MATLAB and Simulink in order to demonstrate practical test scenarios for surrogate safety analysis of vulnerable road users (VRU)-vehicle conflicts at urban roads. Different scenarios are built such as a pedestrian walking and running in front of the automated vehicle from the nearside. In addition to that, uncertain weather conditions and sensor failure are modelled and analysed. The results showed a high impact of weather conditions and sensor failure to the safety measures during traffic conflict. The outcomes reveal that the physics based safety model can mimic the real world scenarios and can support safety analysis in an accurate and cost-effective way.

Swarm technology provides a new opportunity for sensor systems in which the movement of the agents can be leveraged to enhance the joint capacity of the sensors and subsequent signal processing algorithms. In the Localizing Urban Swarm Technology (LocUST) project for DARPA's OFFensive Swarm-Enabled Tactics (OFFSET) program, the authors developed a complex fading model for the virtual radio frequency (RF) environment, decentralized search and localization tasking, and movement-enhanced time difference of arrival (TDOA) localization. The system was also implemented in hardware using Bluetooth 5.0 modules. This paper reports on the fading model, localization algorithm, and hardware testing results, with the companion paper reporting on the swarm coordination, localization-enhancing movement, and associated experimentation.

Swarms of inexpensive, robotic sensors have the potential for revolutionizing intelligence gathering. They self-organize to provide wide apertures, redundancy, attritability, with low probability of detect over a wide area. Coordinating swarm behaviors to provide the necessary apertures and spatial configurations requires novel methods of distributed control that can maintain the positioning accuracy in the face of arbitrary threats and obstacles. In this paper we describe the algorithms to control a swarm of air vehicles with radio frequency receivers that cooperatively search an urban area for radio frequency emitters, self-organize into teams to localize each emitter, and perform coordinated maneuvers to maximize the information gain during the localization operation. The swarm is able to adapt to attrition, performs collision avoidance, and adjusts its trajectories based on the urban terrain. These behaviors were implemented in a ROS-based swarm deployment environment suitable for execution on a small drone and simulated in a 3D model of a small urban area. This paper describes the search and localization tactics employed, the algorithms for implementing those tactics in the swarm, and experimental results. Our companion paper describes the algorithms used for localization.

In contrast of RF jamming devices for counter-UAV measure, which proves ineffective for autonomous UAV, we propose Jellyfish - a vision-based and machine learning driven defensive approach using inexpensive drones. Jellyfish minimizes cost while maximizing the effectiveness of countering autonomous UAV threats. In software aspect, we collect synthetic and real-world imagery and train deep learning model for real-time adversarial drone detection and tracking and then feed tracking output to a highly responsive controller to mirror the movement of the adversary. On the other hand, we design a lightweight and self-releasable capturing mechanism, similar to the tentacles of a jellyfish (hence the name “Jellyfish"), to optimize both the capturing area and defending drone's maneuverability and survivability.

In multinational defence operations, either EU or NATO driven, the exchange of surveillance and reconnaissance data and information is an essential aspect to provide to the commander the needed situational awareness. This improvement of situational awareness, especially in a maritime environment, may be achieved amongst others by extending the ISTAR performance through using unmanned systems (UxS) and integrating them into the combat management system (CMS), ensuring interoperability between the deployed forces and building the overall system based on a solid architecture. Within this frame, the OCEAN2020 (Open Cooperation for European mAritime awareNess) project, funded by the European Union's Preparatory Action on Defence Research and implemented by the European Defence Agency, sees 42 partners from 15 EU countries working together to build future maritime surveillance by integrating drones, unmanned vessels and unmanned submarines into fleet operations. Data and information will be integrated in a comprehensive (maritime) picture of developing situations, enhancing the situational awareness, and thus supporting military commanders on different unit levels in their decision making. This paper focuses on the challenge to define flexible architectures for maritime operations. The reference architecture for a system-of-systems that aims to provide enhanced situational awareness in a naval environment will be presented. The reference architecture provides reusable structures and rules, helping to reduce development and system realization time and costs. This Reference Architecture will establish strategic decisions regarding system technologies to be used and will serve as baseline for the development of two Target Architectures for the two planned demonstrations.

Multiple defense-relevant open architecture standards include the publish/subscribe messaging paradigm, which allows for dynamic network topology and scalability. Using the Transport Layer Security (TLS) protocol to secure such messaging is common; however, certificate validation must be performed. Typically, certificate validation is left to the application to configure, but history has shown that application developers often get incorrect certificate validation. In this paper, we explore the overhead costs of different security implementations under varying network conditions within a pub/sub system. Furthermore, we study how TrustBase strengthens and simplifies certificate validation within a pub/sub architecture. TrustBase allows a system administrator or integrator to specify a single certificate validation policy for all applications in the system. This ensures that even if application developers have misconfigured certificate validation, the policy is followed, which we believe could make system accreditation easier. Our study is conducted on a notional system with an Apache ActiveMQ messaging server. Handshake timing data are collected from several publishers and subscribers to understand the overhead resulting from using TLS with and without the TrustBase kernel module active on the system. Our experiments run with different certificate validation strategies including prepositioned public-keys and certificate chaining with a trusted root certificate authority. To our knowledge, we are the first to study TrustBase in an environment that emulates realistic network conditions and a messaging paradigm beyond the traditional client/server model. Our results confirm those of the original TrustBase work; TrustBase adds negligible overhead and is easily configurable as a universal certificate validation authority.

Experts and novices differ with respect to the use of intuition and deliberation in their decision-making processes, which affects the quality of decisions they make. We often ask ourselves: where does expert intuition and deliberation come from? If the progression from novice to expert is made through learning experiences, what should we provide novices during training? Identifying the discrepancies between experts and novices is essential for developing a computational learning system that simulates the human decision-making process. In this paper, we investigate the difference between individuals operating Unmanned Aerial Vehicle (UAV) missions, collected in a dataset called the Supervisory Control Operations User Test bed (SCOUT), by analyzing their computational models. For the computational models, deep neural networks (DNNs) and double transition models (DTMs) were employed. A set of DNNs was constructed from biometric information about eye movements, and a set of DTMs was built from event-driven data associated with actions taken by the individuals. For investigating DNNs, we examined how much improvement was obtained during training and validation, while for DTMs, we concentrated on the reward distributions of trajectories derived through inverse reinforcement learning (IRL). We classified SCOUT subjects into three levels of expertise according to their selfassessment and the maximum score achieved: novice, intermediate and expert. By analyzing these models, we identified differences between the expert and novice groups. In particular, the accuracy of the expert DNNs improved more effectively than that of novice DNNs, and the reward distributions of the expert DTMs were more closely clustered than those of novice DTMs.

Reinforcement Learning holds the potential to enable many systems with rapid, intelligent automated decision- making. However, reinforcement learning on embodied systems is a much greater challenge than the simulated environments and tasks which have been solved to date. A learner in an embodied system cannot run millions of trials or easily tolerate fatal trajectories. Therefore, the ability to train agents in simulated environments and effectively transfer their knowledge to real-world environments will be crucial, and likely an integral part of constructing future robotic systems. We perform experiments in an original transfer reinforcement learning task we constructed using the game “Sonic 3 and Knuckles," evaluating two transfer learning techniques from the literature.

In this paper, finite horizon intelligent decision-making problem has been investigated for self-organized autonomous systems especially under unstructured environment. According to the latest studies, the uncertainty of environment will seriously affect the effectiveness of decision making especially for autonomous systems. To handle these issues, transfer learning, and deep reinforcement learning has been presented recently. However, those existing Learning algorithms commonly needs a large set of state-space which cause the algorithm to be time-consuming and not suitable for real-time application. Therefore, in this paper, a library of polices trained using Deep Q-Learning under similar environments is built and implemented.

The use of unmanned aerial vehicles (drones) is expanding to commercial, scientific, and agriculture applications, including surveillance, product deliveries and aerial photography. One challenge for applications of drones is detecting obstacles and avoiding collisions. A typical solution to this issue is the use of camera sensors or ultrasonic sensors for obstacle detection or sometimes just manual control (teleoperation). However, these solutions have costs in battery lifetime, payload, operator skill. We note that there will be an air disturbance in the vicinity of the drone when it’s moving close to obstacles or other drones. Our objective is to detect obstacles from monitoring the aforementioned air disturbance, by analyzing the data from the drone’s gyroscope and accelerometer. Results from three experiments using the Crazyflie 2 micro drone are reported here. We show that it is possible to reliably detect when a drone is passing under another using by using data mining algorithms to recognize the air disturbance caused by the other drone.

The drones are nowadays devices able to help human operators in a lot of fields, one of these regards the operations of rescue in disasters or emergency events. In these scenarios, the possibility of using in a rapid way a fleet of drones which can be deployed rapidly and are able to cover a particular disaster area and provide connectivity has high importance. This importance regards the possibility for the drones of permitting the communications between the operators that, otherwise, could have serious difficulties because the current communication technologies heavily rely on the backbone network and the failure of base stations (BSs) due to natural disasters causes communication difficulties for public-safety and emergency communications. The contribution of this work is to explore the use of drones for providing safety communications during natural disasters, where part of the communication infrastructure becomes damaged and dysfunctional. We introduce in the system a human mobility model for disaster events in order to take into account the behavior of the people that in these situations has to move in the area full of obstacles created by the considered disaster. The human mobility affects how to provide connectivity in the area where it is possible to have part of the area most crowded than other. So, the drone that covers this particular region has an overload of traffic and, then it is opportune to redirect the traffic flow in order to guarantee the communication between the operators' devices inside the disaster area.

Drones interconnected in a multi-hop ad-hoc fashion and forming a Flying Ad-Hoc Network (FANET) can be used for accomplishing a lot of different tasks. this context, a key role is played by routing protocols in order to allow data exchange following the best path between source and destination. The contribution of this work is to explore and compare two different typologies of routing protocols such as the classical protocols based on link state or those based on heuristics approach such as bio-inspired (swarm intelligence) ones that make use of real life behaviors in order to resolve complex problems. These new kinds of protocols based on heuristic approach can be lighter and more efficient in comparison with classical protocols and leading to a sub-optimal solution, in a few time and with a more efficient resource consumption remaining always able to resolve in a satisfactory way the specific task.

In this paper, we have analyzed scenarios for leader-follower vehicle convoys that have the potential to be unsafe, hazardous, or even fatal in order to provide insight on dangerous driving conditions for autonomous vehicle platoons. These scenarios were created in a simulation software called Prescan, and are made to reflect the behavior of the vehicle dynamics and sensor characteristics of a real-world vehicle convoy that is currently being tested at the American Center for Mobility (ACM) and sponsored by the Department of Energy (DOE), National Renewable Energy Laboratory (NREL), the U.S. Army Combat Capabilities Development Command - Ground Vehicle Systems Center (U.S. Army CC-DEVCOM) and the Michigan Department of Transportation (MDOT). The follower vehicle in this study is tested under multiple scenarios, equipped with a Dedicated Short Range vehicle-to-vehicle communication system (DSRC), a radar-based adaptive cruise control system (ACC) and a precision global positioning system (GPS).

In this paper develop a novel, quantitative, rigorous and efficient method for risk minimization for control and decision under uncertainty. The crucial components of our approach include a rigorous, efficient risk evaluation method and a stochastic optimization technique. The risk evaluation method is an adaptive Monte Carlo estimation method which is derived from the concept of relative entropy and truncated inverse binomial sampling. The stochastic optimization technique is built upon evolutionary computing methods such as genetic algorithms, where the fitness function is constructed from the adaptive Monte Carlo estimation method. The effectiveness of the proposed method is demonstrated by its applications to the design of PID controllers for uncertain systems, where the probability of performance violation is minimized.

In this paper, we develop a rigorous and efficient method for risk evaluation. Our risk evaluation method is an adaptive Monte Carlo estimation method implemented as a rectangular random walk, which is derived from a mixed error criterion and the concept of relative entropy from information theory. Our proposed method of risk evaluation can be orders of magnitude more efficient as compared to existing methods in literatures and widely used softwares. This new method makes it possible to evaluate risk of systems so that in a strict statistical sense, either the absolute error can be controlled below 10⁻⁶ or the relative error can be controlled below 0.01, that is, the error of risk evaluation can be rigorously certified at extremely low level, which is impossible by using existing methods.

In this paper, we develop adaptive PAC (probably approximately correct) learning methods with applications to design control strategy for uncertain systems. The proposed PAC learning methods mimic the adaptive learning behavior of human being to accumulate evidence step by step and make decisions based on available observations. In the proposed methods, new comparative inferential techniques are developed to quickly eliminate inferior hypotheses. We demonstrate that the proposed PAC learning methods are substantially more efficient in finding the optimal hypothesis with pre-specified level of confidence and accuracy. The proposed PAC learning methods can be applied to the design of robust controllers, where the uncertain parameters of the relevant system is sampled to obtain training examples for the learning process.

A persistence concern of control engineering is the performance of systems in the presence of uncer- tainty. By treating uncertain parameters of systems as random variables, the performance of systems may be formulated as means of random variables. In this paper, we develop multistage schemes for making statistical inference of means of random variables. Such schemes are unprecedentedly e±cient as compared to existing methods, while guaranteed pre-speci¯ed level of credibility. The optimality of the proposed schemes is established by making use of the uniform exponential maximal inequalities. The proposed schemes are applied to robustness analysis of control systems under uncertainty. It is demonstrated the computational complexity of the proposed schemes is substantially lower and inde- pendent of the problem size, as compared to the non-polynomial complexity of the worst-case method of robustness analysis.

Control systems are usually designed based on nominal values of relevant physical parameters. To ensure that a control system will work properly when the relevant physical parameters vary within certain range, it is crucial to investigate how the performance measure affected by the variation of system parameters. In this paper, we demonstrate that such issue boils down to the study of the variation of functions of uncertainty. Motivated by this vision, we propose a general theory for inferring function of uncertainties. By virtue of such theory, we investigate concentration phenomenon of bounded random vectors. We derive multidimensional concentration inequalities for bounded random vectors, which are substantially tighter as compared to existing ones. The new concentration inequalities are applied to investigate the performance of control systems with real parametric uncertainty. It is demonstrated much more useful insights of control systems can be obtained. Moreover, the concentration inequalities offer performance analysis in a significantly less conservative way as compared to the classical deterministic worst-case method.

As the world develops, new and more advanced ways of transportation are invented; i.e. drones. Drones are used in several applications. However, the drone market does not utilize the need of medical emergency drones today, where these drones can be used to save countless lives in severe cases, e.g. sudden cardiac arrest. In case of cardiac arrest, defibrillators may save the life if it reaches the victims within short time. It raises the survival rate exponentially. Nonetheless, reaching the victims in a short period of time is challenging as the weight of the equipment is large. This work aims to design an autonomous drone that will be able to carry heavy payloads (portable medical equipment) while being fast and agile. The medical equipment/components are studied to choose the most fit for the proposed design in terms of efficiency and weight. The drone’s components are compared and studied in detail, allowing to choose the fittest motors, ESCs, frame, battery, and propellers. After which the quadcopter’s ability is expected to successfully achieve the objective of trying to save victim life in the city of Sharjah. In addition, the work includes a SolidWorks analysis to the design of the drone’s mechanical components to estimate the possibility of failure.

This work investigates design of a drone for transporting valuable objects. All the required components for building the drone and discussed. The drone can carry a maximum payload of 10 kg for a long duration over reasonable distance. This is achieved by using a hybrid mechanism that combines two 25cc fuel engines with four electrical motors operating by two 6S lithium polymer (LiPo) batteries. The hybrid mechanism is chosen as it tackles the electric drone main problems which are: short flight duration and low payload-carrying ability.