Browse Topic: Machine learning

Items (994)
Manual installation of vehicle underbody Grommets is tedious task which sometimes results in incomplete & inaccurate installation. Based on process quality guidelines, this comes under potential defect category at End of line inspection area for which few hours of manual efforts are required for identifying location of error & doing rework activity. Existing deep learning & image comparison method falls short in identifying error location. To overcome this challenge, deep learning algorithm with co-ordinate system developed which comprises of identifying class or category of entities present in test and reference image and indexing it as Grommet or Hole. This further comprises of determining 2D position in terms of X and Y co-ordinate of each indexed hole & Grommet between reference & test image. This approach results in precise comparison & identification of error in terms of missing or misplacing of Grommets which also offers significant saving in manual installation and rework
Dhumal, Abhishek TrimbakMishra, JagdishTote, AnujNurukurthi, LakshmiKumar, Prakash
Looking at the current scenario in transportation industry, in majority of the conventional powertrains, internal combustion (IC) engines fueled by diesel serve as the powerhouse. In all locomotives driven by IC engine, it is essential to monitor critical engine parameters to ensure good engine health and performance. Exhaust temperature of engine is a very critical parameter which gives the information about in-cylinder combustion. In traditional diesel engine layouts, exhaust temperature measurement relies on physical temperature sensor. The proposed methodology is focused on applying the data driven models for providing an estimated value of the exhaust temperature. Based on the estimated value of exhaust temperature, this technique can be used to monitor the IC engine. This methodology uses an advanced Artificial Intelligence technique for providing an accurate estimate of exhaust gas temperature. Real world vehicle data was used for training, validating, and testing the data
Jagtap, Virendra ShashikantGanguly, GouravMitra, Partha
In Automotive world, vehicle development includes design and testing of hardware and software. Hardware includes components required for actuation and sensing, along with the controller hardware. Software includes control logic embedded in controller for functioning of these components. Generally, software inside controller could be validated in various ways e.g., Software in Loop (SiL), Hardware in Loop (HiL), Vehicle testing. During initial phase of control software development cycle, plant models with adequate accuracy replicating hardware components are utilized for digital software validation. Many a times, hardware components might be available before control software matures. Hence, to validate plant models for their accuracy & quality alternate option of actual controller is needed during initial phase. Intelligent controller mimicking original controller can be an alternate option for plant model improvement and component level performance analysis. This paper proposes a
Chhagar, Rohnit SinghNavse, SiddharthKumar, Lavanya
As vehicle emission standards are becoming stringent worldwide because of the looming climate crisis, it is important to control the pollutants that vehicles emit. To achieve the stringent emission target, it has become a priority to enhance the capability of Emission Control System (ECS) which consist of Diesel Oxidation Catalyst (DOC), Diesel Particulate Filter (DPF) and Selective Catalytic Reduction (SCR) sub-systems. One of the bottlenecks is the limited operating temperature range of the after-treatment system. In modern emission control systems, the temperature characteristics should always be optimized to have the best efficiency involving chemical conversions. To achieve this optimal operating temperature, different thermal control strategies are followed in the Engine and emission control unit. Temperature sensor values are one of the primary inputs for thermal management strategies. In the event of temperature sensor malfunction, the ECS performance is affected due to
Kumar, AmitV H, YashwanthKumar, RamanHegde, KarthikManojdharan, Arjungopal
The new Bharat Stage (BS) VI Stage 2 regulation for automotive vehicles in India requires monitoring the performance of emission control components, such as Selective Catalytic Reduction (SCR) systems, Diesel Oxidation Catalysts (DOCs), Diesel Particulate Filter (DPF), Nitrous Oxides (NOx) Sensors, and Exhaust Gas Recirculation (EGR). The regulation also mandates that a minimum In-Use Performance Ratio (IUPR) must be met, which is the ratio of the number of times a component's performance is monitored to the number of drive cycles the engine has undergone. The IUPR must be tracked throughout the vehicle's lifetime after an initial run-in period. In an effort comply with the minimum IUPR requirement, the engine and after-treatment system calibrations must ensure that the conditions and threshold ranges for enabling performance monitoring of emission-critical components are met across all vehicles operating duty cycles and varying geographic conditions. This study explores the novel
Kumar, KosalaramanVenkat, HarishAvanashilingam, Jayanth Balaji
A BDT (Battery digital Twin) is a virtual representation of a vehicle's physical battery system, combining electrochemical and machine learning models to provide insights into key battery parameters like State of Charge (SOC), State of Health (SOH), Internal Resistance (IR), and Remaining Useful Life (RUL). This BDT model is calibrated using cell testing throughout its degradation process up to 80% SOH, alongside vehicle data for accurate predictions under diverse conditions. By continuously monitoring the battery under various operating scenarios, the BDT aids in effective battery management, identifying cells that degrade more quickly and the likely causes of this degradation. Current and temperature profiles offer insights into battery usage patterns. The BDT aggregates fleet-wide parameters and analyzes individual cell performance, providing critical information on SOC, SOH, IR, RUL, and voltage. Additionally, the BDT includes prognostic capabilities to alert users of potential
Sasi Kiran, TalabhaktulaKondhare, ManishPatil, SuyogNath, SubhrajyotiCH, Sri RamTank, PrabhuSarkar, Prasanta
Dynamic Vehicle mass is one of the most critical parameters in automotive controls such as battery management, transmission shift scheduling, distance-to-empty predictions and most importantly, various active and passive safety systems. This work aims to find out dynamic Vehicle mass for Electric Vehicles in real time transient driving conditions. The work proposes a real-time approach in finding Dynamic vehicle mass where accumulated Energy based vehicle performance, an improvement to the vehicle dynamics equation, has been employed for consistent and accurate results. Factors affecting vehicle mass such as road grade, dynamic friction coefficient, driving pattern, wheel slip etc. have been considered for model optimization. Here recursive Bayesian state estimator has been used for finding vehicle mass as a constant state variable while time varying forgetting factors are used to nullify the impact of major losses. Algorithm is auto tuned using Machine Learning techniques to first
Pandey, SuchitSarkar, PrasantaSawhney, ChandanKondhare, ManishJoshi, PawanCH, Sri Ram
Cybersecurity, particularly in the automotive sector, is of paramount importance in today’s digital age. With the advent of connected commercial vehicles, which leverage telematics for efficient fleet management, the landscape of automotive cybersecurity is rapidly evolving. These vehicles, integral to logistics and transportation businesses, are becoming increasingly connected, thereby escalating the risks associated with cybersecurity threats. These commercial vehicles are becoming prime targets for cyber-attacks due to their connectivity and the valuable data they hold. The potential consequences of these cyber-attacks can range from data breaches to disruptions in fleet operations, and even safety risks. This paper analyses the unique challenges faced by the commercial vehicle sector, such as the need for robust telematics systems, secure communication channels, and stringent data protection measures. Case studies of notable cybersecurity incidents involving commercial vehicles are
Mahendrakar, ShrinidhiMadarla, ManojGangapuram, SivaDadoo, Vishal
This study presents a comprehensive survey of the current state-of-the-art techniques in virtual scene generation, particularly within the context of autonomous driving. The integration of deep learning methods such as generative adversarial networks (GANs) and convolutional LSTM (ConvLSTM) is explored in detail. Additionally, the effectiveness and applicability of these techniques in simulating real-world traffic scenarios are analyzed. Our article aims to bridge the gap between theoretical models and practical applications, providing an in-depth understanding of how deep learning and virtual scene generation converge to enhance the efficacy of autonomous driving systems
Ayyildiz, Dilara VefaAlnaser, Ala JamilTaj, ShahramZakaria, MahtaJaimes, Luis Gabriel
ABSTRACT In this paper we present an intelligent power controller for a vehicle power system that employs multiple power sources. In particular we focus on a vehicle power system architecture that is used in vehicles such as Mine Resistant Ambush Protected (MRAP) vehicle. These vehicles are designed to survive IED (Improvised Explosive Devices) attacks and ambushes. The power system has the following major components: a “clean” bus, a “dirty” bus, an engine, a hydraulic system and a switch between the clean and the dirty bus. We developed algorithms for intelligent energy management for this type of vehicle power system including DP (Dynamic Programming) optimization, DP online control and a machine learning technique that combines neural networks with DP to train an intelligent power controller. We present experiments conducted through modeling and simulation using a generic commercial software tool and a lab hardware setup
Chen, ZhihangMurphey, Yi L.Chen, ZhengMasrur, AbulMi, Chris
ABSTRACT A promising approach to autonomous driving is machine learning. In machine learning systems, training datasets are created that capture the sensory input to a vehicle as well as the desired response. One disadvantage of using a learned navigation system is that the learning process itself may require both a huge number of training examples and a large amount of computing. To avoid the need to collect a large training set of driving examples, we describe a system that takes advantage of the immense number of training examples provided by ImageNet, but at the same time is able to adapt quickly using a small training set for the driving environment
Provodin, ArtemTorabi, LiilaMuller, UrsFlepp, BeatSergio, MichaelŽbontar, JureLeCun, YannJackel, L. D.
ABSTRACT Timely part procurement is vital to the maintenance and performance of deployed military equipment. Yet, logistical hurdles can delay this process, which can compromise efficiency and mission success for the warfighter. Point-of-need part procurement through additive manufacturing (AM) is a means to circumvent these logistical challenges. An Integrated Computational Materials Engineering framework is presented as a means to validate and quantify the performance of AM replacement parts. Statistical modeling using a random forest network and finite element modeling were to inform the build design. Validation was performed by testing coupons extracted from each legacy replacement parts, as well as the new additively manufactured replacement parts through monotonic tensile and combined tension-torsion fatigue testing. Destructive full hinge assembly tests were also performed as part of the experimental characterization. Lastly, the collected experimental results were used to
Gallmeyer, Thomas GDahal, JineshKappes, Branden BStebner, Aaron PThyagarajan, Ravi SMiranda, Juan APilchak, AdamNuechterlein, Jacob
ABSTRACT Recurrent Neural Networks have largely been explored for low-dimensional time-series tasks due to their fading memory properties, which is not needed for feed-forward methods like the Convolutional Neural Network. However, benefits of using a recurrent-based neural network (i.e. reservoir computing) for time-independent inputs includes faster training times, lower training requirements, and reduced computational burdens, along with competitive performances to standard machine learning methods. This is especially important for high-dimensional signals like complex images. In this report, a modified Echo State Network (ESN) is introduced and evaluated for its ability to perform semantic segmentation. The parallel ESN containing 16 parallel reservoirs has an image processing time of 2 seconds with an 88% classification rate of 3 classes, with no prior feature extraction or normalization, and a training time of under 2 minutes. Citation: S. Gardner, M. R. Haider, J. Smereka, P
Gardner, S.Haider, M.R.Smereka, J.Jayakumar, P.Kulkarni, K.Gorsich, D.Moradi, L.Vantsevich, V.
ABSTRACT Future autonomous combat vehicles will need to travel off-road through poorly mapped environments. Three-dimensional topography may be known only to a limited extent (e.g. coarse height), but this will likely be noisy and of limited resolution. For ground vehicles, 3D topography will impact how far ahead the vehicle can “see”. Higher vantage points and clear views provide much more useful path planning data than lower vantage points and occluded views from trees and structures. The challenge is incorporating this knowledge into a path planning solution. When should the robot climb higher to get a better view or else continue moving along the shortest path predicted by current information? We investigated the use of Deep Q-Networks (DQN) to reason over this decision space, comparing performance to conventional methods. In the presence of significant sensor noise, the DQN was more successful in finding a path to the target than A* for all but one type of terrain. Citation: E
Martinson, EricPurman, BenDallas, Andy
ABSTRACT An increasing pace of technology advancements and recent heavy investment by potential adversaries has eroded the Army’s overmatch and spurred significant changes to the modernization enterprise. Commercial ground vehicle industry solutions are not directly applicable to Army acquisitions because of volume, usage and life cycle requirement differences. In order to meet increasingly aggressive schedule goals while ensuring high quality materiel, the Army acquisition and test and evaluation communities need to retain flexibility and continue to pursue novel analytic methods. Fully utilizing test and field data and incorporating advanced techniques, such as, big data analytics and machine learning can lead to smarter, more rapid acquisition and a better overall product for the Soldier. Logistics data collections during operationally relevant events that were originally intended for the development of condition based maintenance procedures in particular have been shown to provide
Heine, RichardFrounfelker, BradSalins, LaneWang, Chongying
ABSTRACT With recent advancements in the automotive world and the introductions of autonomous vehicles, automotive cybersecurity has become a main and primary issue for every automaker. In order to come up with measures to detect and protect against malicious attacks, intrusion detection systems (IDS) are commonly used. These systems identify attacks while comparing normal behavior with abnormalities. In this paper, we propose a novel, two-stage IDS based on deep-learning and rule-based systems. The objective of this IDS is to detect malicious attacks and ensure CAN security in real time. Deep Learning has already been used in CAN IDS and is already proven to be a successful algorithm when it comes to extensive datasets but comes with the cost of high computational requirements. The novelty of this paper is to use Deep Learning to achieve high predictability results while keeping low computational requirements by offsetting it with rule-based systems. In addition, we examine the
Zhang, LinxiKaja, NevrusShi, LyndonMa, Di
ABSTRACT The fundamental aspect of unmanned ground vehicle (UGV) navigation, especially over off-road environments, are representations of terrain describing geometry, types, and traversability. One of the typical representations of the environment is digital surface models (DSMs) which efficiently encode geometric information. In this research, we propose a collaborative approach for UGV navigation through unmanned aerial vehicle (UAV) mapping to create semantic DSMs, by leveraging the UAV wide field of view and nadir perspective for map surveying. Semantic segmentation models for terrain recognition are affected by sensing modality as well as dataset availability. We explored and developed semantic segmentation deep convolutional neural networks (CNN) models to construct semantic DSMs. We further conducted a thorough quantitative and qualitative analysis regarding image modalities (between RGB, RGB+DSM and RG+DSM) and dataset availability effects on the performance of segmentation
Brand, Howard J. J.Li, Bing
ABSTRACT Simulation is a critical step in the development of autonomous systems. This paper outlines the development and use of a dynamically linked library for the Mississippi State University Autonomous Vehicle Simulator (MAVS). The MAVS is a library of simulation tools designed to allow for real-time, high performance, ray traced simulation capabilities for off-road autonomous vehicles. It includes features such as automated off-road terrain generation, automatic data labeling for camera and LIDAR, and swappable vehicle dynamics models. Many machine learning tools today leverage Python for development. To use these tools and provide an easy to use interface, Python bindings were developed for the MAVS. The need for these bindings and their implementation is described. Citation: C. Hudson, C. Goodin, Z. Miller, W. Wheeler, D. Carruth, “Mississippi State University Autonomous Vehicle Simulation Library”, In Proceedings of the Ground Vehicle Systems Engineering and Technology Symposium
Hudson, Christopher R.Goodin, ChristopherMiller, ZachWheeler, WarrenCarruth, Daniel W.
ABSTRACT Recent advances in neuroscience, signal processing, machine learning, and related technologies have made it possible to reliably detect brain signatures specific to visual target recognition in real time. Utilizing these technologies together has shown an increase in the speed and accuracy of visual target identification over traditional visual scanning techniques. Images containing a target of interest elicit a unique neural signature in the brain (e.g. P300 event-related potential) when detected by the human observer. Computer vision exploits the P300-based signal to identify specific features in the target image that are different from other non-target images. Coupling the brain and computer in this way along with using rapid serial visual presentation (RSVP) of the images enables large image datasets to be accurately interrogated in a short amount of time. Together this technology allows for potential military applications ranging from image triaging for the image analyst
Ries, Anthony J.Lance, BrentSajda, Paul
ABSTRACT Motion planning algorithms for vehicles in an offroad environment have to contend with the significant vertical motion induced by the uneven terrain. Besides the obvious problems related to driver comfort, for autonomous vehicles, such “bumpy” vertical motion can induce significant mechanical noise in the real time data acquired from onboard sensors such as cameras to the point that perception becomes especially challenging. This paper advances a framework to address the problem of vertical motion in offroad autonomous motion control for vehicular systems. This framework is first developed to demonstrate the stabilization of the sprung mass in a modified quarter-car tracking a desired velocity while traversing a terrain with changing height. Even for an idealized model such as the quarter-car the dynamics turn out to be nonlinear and a model-based controller is not obvious. We therefore formulate this control problem as a Markov decision process and solve it using deep
Salvi, AmeyaBuzhardt, JakeTallapragada, PhanindraKrovi, VenkatBrudnak, MarkSmereka, Jonathon M.
ABSTRACT This paper describes the use of neural networks to enhance simulations for subsequent training of anomaly-detection systems. Simulations can provide edge conditions for anomaly detection which may be sparse or non-existent in real-world data. Simulations suffer, however, by producing data that is “too clean” resulting in anomaly detection systems that cannot transition from simulated data to actual conditions. Our approach enhances simulations using neural networks trained on real-world data to create outputs that are more realistic and variable than traditional simulations. Citation: P.Feldman, “Training robust anomaly detection using ML-Enhanced simulations”, In Proceedings of the Ground Vehicle Systems Engineering and Technology Symposium (GVSETS), NDIA, Novi, MI, Aug. 11-13, 2020
Feldman, Philip
ABSTRACT This paper will explore the opportunities for artificial intelligence (AI) in the system engineering domain, particularly in ways that unite the unique capabilities of the systems engineer with the AI. This collaboration of human and machine intelligence is known as Augmented Intelligence (AuI). There is little doubt that systems engineering productivity could be improved with effective utilization of well-established AI techniques, such as machine learning, natural language processing, and statistical models. However, human engineers excel at many tasks that remain difficult for AIs, such as visual interpretation, abstract pattern matching, and drawing broad inferences based on experience. Combining the best of AI and human capabilities, along with effective human/machine interactions and data visualization, offers the potential for orders-of-magnitude improvements in the speed and quality of delivered
Petrotta, MarkPeterson, Troy
ABSTRACT Crowdsourcing is an overarching term that denotes a number of ways to use the web as means to enlist a large number of individuals to perform a particular task. The tasks can range from simply providing an opinion, to contributing material, to solving a problem. Because the term crowdsourcing is used to denote a variety of activities in many different contexts, strong opinions have formed in many minds. This paper is an attempt to inform the reader of the complexity that underlies the simple term “crowdsourcing.” We then describe the connection between the DARPA Adaptive Vehicle Make program with the potential limitations of crowdsourcing complex tasks using examples from industry. Using these examples, we present a research motivation detailing areas to be improved within current crowdsourcing frameworks. Finally, an agent-based simulation using machine learning techniques is defined, preliminary results are presented, and future research directions are described
Gerth, Richard J.Burnap, AlexPapalambros, Panos
Summary Growing environmental concerns coupled with the complex issue of global crude oil supplies drive automobile industry towards the development of fuel-efficient vehicles. Due to the possible multiple-power-source nature and the complex configuration and operation modes, the control strategy of a military vehicle is more complicated than that of a conventional vehicle. Furthermore, military vehicles often have heavier weights and are used to operate multiple functions such as engaging weapons, turning on sensors, silent watch, etc., which results in big load fluctuation. In this paper we present our research in optimizing power flow in a heavy vehicle for a given mission plan. A mission plan consists of a sequence of operations and speed profiles. The vehicle architecture will be modeled based on Stryker power system which consists of a diesel engine, a main battery pack, an auxiliary battery pack, and an APU. The APU can supply power to the auxiliary loads and auxiliary batteries
Murphey, Yi L.Masrur, M. AbulNeumann, Donald E.
ABSTRACT Knowing the soil’s strength properties is a vital component to accurately develop Go/No-Go mobility maps for the Next Generation NATO Reference Mobility Model (NG-NRMM). The Unified Soil Classification System (USCS) and soil strength of the top 0-6” and 6-12” of the soil are essential terrain inputs for the model. Current methods for the NG-NRMM require in-situ measurement of soil strength using a bevameter, cone penetrometer, or other mechanical contact device. This study examines the use of hyperspectral and thermal imagery to provide ways of remotely characterizing soil type and strength. Hyperspectral imaging provides unique spectrums for each soil where a Soil Classification Index (SCI) was developed to predict the gradation of the soil types. This gradation provides a means of identifying the soil type via the major divisions within the USCS classification system. Thermal imagery is utilized to collect the Apparent Thermal Inertia (ATI) for each pit, which is then
Ewing, JordanOommen, ThomasJayakumar, ParamsothyAlger, Russell
ABSTRACT The Soar Cognitive Architecture is a reasoning system that enables knowledge-rich, mission focused reasoning including integration of bottom-up, sensor-driven reasoning and top-down, context-driven reasoning, and more intelligent use of existing sensors. This reasoning is a combination of deliberate (e.g., planning) and reactive (e.g., hard-coded) behaviors. We are applying Soar on a current effort to (1) increase autonomy and (2) achieve equivalent or superior performance while controlling weight, energy, and costs
Marinier, RobertBechtel, RobertDallas, Andrew
ABSTRACT Can convolutional neural networks (CNNs) recognize gestures from a camera for robotic control? We examine this question using a small set of vehicle control gestures (move forward, grab control, no gesture, release control, stop, turn left, and turn right). Deep learning methods typically require large amounts of training data. For image recognition, the ImageNet data set is a widely used data set that consists of millions of labeled images. We do not expect to be able to collect a similar volume of training data for vehicle control gestures. Our method applies transfer learning to initialize the weights of the convolutional layers of the CNN to values obtained through training on the ImageNet data set. The fully connected layers of our network are then trained on a smaller set of gesture data that we collected and labeled. Our data set consists of about 50,000 images recorded at ten frames per second, collected and labeled in less than 15 man-hours. Images contain multiple
Kawatsu, ChrisKoss, FrankGillies, AndyZhao, AaronCrossman, JacobPurman, BenStone, DaveDahn, Dawn
ABSTRACT Bayesian networks have been applied to many different domains in order to perform prognostics, reduce risk and ultimately improve decision making. However, these methods have not been applied to military ground vehicle field data sets. The primary objective of this study is to illustrate how Bayesian networks can be applied to a ground vehicle data set in order to predict potential downtime. The study generated a representative field data set, along with tabu search, in order to learn the network structure followed by quantification of link probabilities. The method is illustrated in a case study and future work is described in order to integrate the method into a real-time monitoring system. The study yielded a highly accurate prediction algorithm that can improve decision making, reduce downtime and more efficiently manage resources in the ground vehicle community
Banghart, MarcNelson, DavidBrennan, Adam
ABSTRACT Modern perception systems for autonomous vehicles are often dependent on deep neural networks, however, such networks are unfortunately susceptible to subtle perturbations to their inputs. Due to the interconnected nature of perception/control systems in autonomous vehicles, it is quite difficult to evaluate the autonomy stack’s robustness in different scenarios. Numerous tools have been developed to assist developers increase the robustness of these algorithms for on-road driving, but little has been accomplished for off-road driving. This work aims to bridge this gap by presenting a reinforcement learning framework to identify unsuspecting off-road scenes that confuse a custom autonomy stack with a DNN-based perception algorithm to ultimately lead the vehicle into a collision. Citation: T. Sender, M. Brudnak, R. Steiger, R. Vasudevan, B. Epureanu, “Using Deep Reinforcement Learning to Generate Adversarial Scenarios for Off-Road Autonomous Vehicles,” In Proceedings of the
Sender, TedBrudnak, MarkSteiger, ReidVasudevan, RamEpureanu, Bogdan
ABSTRACT The data-driven machine learning (ML) method is developed to rapidly evaluate the thermal and flow fields of a ground vehicle and its neighboring environment at various conditions. The artificial neural network (ANN) is implemented as the ML model to evaluate the fields, while achieving equivalent accuracy as the CFD simulations. In order for ANN to precisely map a relationship between the simulation parameters and the solution field, the proper orthogonal decomposition (POD) technique is applied to reduce the dimension of the field variables. Consequently, the compressed data (i.e. modal coefficients) is selected as the target for the ANN. Once trained, POD reconstruction is performed on the ANN predicted modal coefficients to recover the CFD solution. The developed framework is tested at diverse sample sites, and the maximum mean absolute errors are found to be 0.41 K and 0.019 m/s for thermal and flow simulations, respectively, verifying the outstanding prediction
Hong, Seong HyeonHouse, AlecKaminsky, Andrew L.Tison, NathanRuan, YeefengKorivi, VamshiWang, YiPant, Kapil
ABSTRACT We describe a simulation environment that enables the design and testing of control policies for off-road mobility of autonomous agents. The environment is demonstrated in conjunction with the design and assessment of a reinforcement learning policy that uses sensor fusion and inter-agent communication to enable the movement of mixed convoys of conventional and autonomous vehicles. Policies learned on rigid terrain are shown to transfer to hard (silt-like) and soft (snow-like) deformable terrains. The enabling simulation environment, which is Chrono-centric, is used as follows: the training occurs in the GymChrono learning environment using PyChrono, the Python interface to Chrono. The GymChrono-generated policy is subsequently deployed for testing in SynChrono, a scalable, cluster-deployable multi-agent testing infrastructure that uses MPI. The Chrono::Sensor module simulates sensing channels used in the learning and inference processes. The software stack described is open
Negrut, D.Serban, R.Elmquist, A.Taves, J.Young, A.Tasora, A.Benatti, S.
ABSTRACT Determining where a vehicle can and cannot safely drive is a fundamental problem that must be answered for all types of vehicle automation. This problem is more challenging in cold regions. Trafficability characteristics of snow and ice surfaces can vary greatly due to factors such as snow depth, strength, density, and friction characteristics. Current technologies do not detect the type of snow or ice surface and therefore do not adequately predict trafficability of these surfaces. In this paper, we took a first step towards developing a machine vision classifier with an exploratory analysis and classification of cold regions surface images. Specifically, we aimed to discriminate between packed snow, virgin snow, and ice surfaces using a series of classical machine learning and deep learning methods. To train the classifiers, we captured photographs of surfaces in real world environments alongside hyperspectral scans, spectral reflectance measurements, and LIDAR. In this
Welling, OrianMeyer, AaronVecherin, SergeyParker, Michael
ABSTRACT Analytical performance assessment of Active Protection Systems (APS) and the vulnerability assessment of ground vehicles using classical physics-based modeling and simulations has many challenges. Also, modeling many of the factors involved in the interaction during Hard-Kill (HK) of the incoming threat with a countermeasure and the resulting outcomes are quite complex and have varied effects on the survivability of the vehicle. Therefore, relying only on deterministic solutions, are time consuming and computationally cost prohibitive. This effort is focused on changing this paradigm by researching for a suitable machine learning algorithm which takes in simulation data from high fidelity physics-based models as training data. Through decomposition, interpolation and reconstruction techniques, surrogate models can be constructed using the simulation data. These surrogate models can then be used for a quick assessment (fraction of a second compared to a day per simulation
Kulkarni, Kumar BBabu, VenkateshKankanalapalli, SanjayVipradas, AdityaJayakumar, PThyagarajan, Ravi
ABSTRACT This paper describes research into the applicability of anomaly detection algorithms using machine learning and time-magnitude thresholding to determine when an autonomous vehicle sensor network has been subjected to a cyber-attack or sensor error. While the research community has been active in autonomous vehicle vulnerability exploitation, there are often no well-established solutions to address these threats. In order to better address the lag, it is necessary to develop generalizable solutions which can be applied broadly across a variety of vehicle sensors. The current measured results achieved for time-magnitude thresholding during this research shows a promising aptitude for anomaly detection on direct sensor data in autonomous vehicle platforms. The results of this research can lead to a solution that fully addresses concerns of cyber-security and information assurance in autonomous vehicles. Citation: R. McBee, J. Wolford, A. Garza, “Detection and Mitigation of
McBee, RyanWolford, JonathanGarza, Abe
ABSTRACT Due to the high complexity of modern internal combustion engines and powertrain systems, the proper calibration of the electronic control unit’s (ECU) parameters has a strong impact on project targets like fuel consumption, emissions and drivability, as well as development costs and project duration. Simulation methods representing the system behavior with a model can support the calibration process considerably. However, standard physics-based models are often not able to describe all effects with sufficient accuracy, or the effort to set up a detailed model is too high. Physics-based models can also have a high computational demand, so that their simulation is not real-time capable. More suited for ECU calibration are data-driven models, combined with Design of Experiment (DoE). The system to be calibrated is identified with few specific test bench or vehicle measurements. From these measurements, a mathematical regression model is built. This paper describes recently
Gutjahr, TobiasKruse, ThomasHuber, Thorsten
ABSTRACT Traditionally, the life cycle management of military vehicle fleets is a lengthy and costly process involving maintenance crews completing numerous and oftentimes unnecessary inspections and diagnostics tests. Recent technological advances have allowed for the automation of life cycle management processes of complex systems. In this paper, we present our process for applying artificial intelligence (AI) and machine learning (ML) in the life cycle management of military vehicle fleets, using a Ground Vehicle fleet. We outline the data processing and data mapping methodologies needed for generating AI/ML model training data. We then use AI and ML methods to refine our training sets and labels. Finally, we outline a Random Forest classification model for identifying system failures and associated root causes. Our evaluation of the Random Forest model results show that our approach can predict system failures and associated root causes with 96% accuracy
Kern, Maxwell C.Cengic, Arif
ABSTRACT Machine learning (ML), artificial intelligence (AI), and computational photography (CP) are pushing the boundaries of how electro-optical (EO) and infra-red (IR) sensors are being used. Especially within military environments, users are asking much more from EO and IR sensor suites. While hardware capability continues to advance the state of the art, software has become the true differentiator for how well these sensor platforms perform for the warfighter. This paper presents work that Consolidated Resource Imaging (CRI) has been developing in the areas of machine learning and computational photography. In this effort, we will discuss two areas of understanding: imagery meant for machine vision and imagery meant for human consumption. We will show how the intersection of machine learning and computational photography allow the symbiotic relationship between the human and the computer. Citation: A. Paul Skentzos, B. Stephen Pizzo, “Balancing Between Computer and Machine Vision
Skentzos, PaulPizzo, Stephen
ABSTRACT We present the results of an exploratory investigation of applying a hybrid quantum-classical architecture to an off-road vehicle mobility problem, namely the generation of go/no-go maps posed as a machine learning problem. The premise of this work rests on two observations. First, quantum computing allows in principle for algorithms that provide a speedup over the best known classical counterparts. However, as it is to be expected of such novel and complex tools (both hardware and algorithmic) at this early developmental stage, current quantum algorithms do not always perform well on real-world problems. Second, complex physics-based vehicle and terramechanics models and simulations, currently advocated for high-fidelity high-accuracy ground vehicle–terrain interaction analyses, pose significant computational burden, especially when applied to mobility studies which may require numerous simulation runs. We describe the Quantum-Assisted Helmholtz Machine formulation, suitable
Serban, RaduWilson, MaxBenedetti, MarcelloRealpe-Gómez, JohnPerdomo-Ortiz, AlejandroPetukhov, AndreJayakumar, Paramsothy
Items per page:
1 – 50 of 994