2022 ARC Research Seminar - Winter Series

PI: Dr. Cameron J. Turner, Co-PI: Dr. Margaret Wiecek, Graduate Students: Philip de Castro and Hannah Stewart (Clemson University); Greg Hartman, Denise Rizzo, David Gorsich, Annette Skowronska, Rachel Agusti (US Army GVSC)
Link to project

Abstract: Tradespace analysis is conducted with the purpose of establishing acceptable target values amongst a set of functional objectives. Through a sequential decision-making process, tradeoffs are identified between objectives to arrive at a set of acceptable target values for each of the functional objectives in the problem. In some cases, this problem spans forty or more objectives, making decision-making difficult. This research effort explores the prospect of evaluating a tradespace by the use of problem decomposition and coordination to reduce the dimensionality of a Pareto optimization problem. In this presentation, we will describe how we built a synthetic tradespace model and applied decomposition and coordination as a decision-making strategy for tradespace optimization. The example problem selected consists of four objectives developed for small mobile semi-autonomous vehicles (such as the Small Military Equipment Transport or SMET and the Deep Orange 13 project vehicle at Clemson). Ongoing work is extending this work to larger decomposition and coordination problems.

Lee Moradi, Haibin Ning, Moataz Khalifa, Vladimir Vantsevich, Samuel Misko, Jesse Paldan, Jordan Whitson (University of Alabama, Birmingham); Yeefend Ruan, Vamshi Korivi (GVSC)
Link to project

Abstract: This project establishes a foundation for advanced tactical and operational mobility controls based on a new technical paradigm of agile tire slippage dynamics that is centered on extremely fast and precise response of the tire to its loading and to transient variations of terrain conditions. For this purpose, new modalities of tire-wheel locomotion have been studied during the first year of the project, including instantaneous tire deflections and strain in three dimensions to characterize the spatial distribution of the tire’s circumferential and radial compressions, instantaneous rolling radii in the driven (at zero torque) and driving (non-zero torque) wheel operational modes, and instantaneous tire slippage.
The seminar presents a new digital image correlation measurement methodology developed for accurate measurement of the above-listed instantaneous tire characteristics. Demonstration of these new techniques is provided based on high-speed video data collected during dynamic testing of a Continental MPT-81 365/80 R20 tire. These measurements are also used to calibrate both the static and dynamic behavior of finite element models of the tire with corresponding boundary and loading conditions applied. The FEM calibrated models will be used during the second year of the project to extrapolate tire behavior in various conditions to inform future research directions for direct in-situ measurement and control of instantaneous tire slippage.

Masood Ghasemi, Vladimir Vantsevich, Lee Moradi, Jesse Paldan (University of Alabama, Birmingham); Jill Goryca, Amandeep Singh (GVSC)
Link to project

Abstract: An in-wheel-motor (IWM) powertrain is a state-of-the-art technology, which allows mobility control, maneuverability, and energy efficiency of vehicles. In this presentation, the control problem of a single-wheel module (SWM) operating in an off-road environment with severe terrain conditions is considered. First, an agile control system is developed in order to achieve precise tracking of a reference angular speed of the wheel. The method integrates a model-free force/torque field estimation with a sliding mode controller to achieve a robust and precise tracking solution. Next, the approach is extended to the linear speed control by incorporating an adjusted reference angular speed. Further, the approach incorporates different slippage suppression methods. Specifically, varying control gains approach, an approach based on correcting the reference angular speed according to a given reference slippage, and a fuzzy logic approach, which addresses a range of characteristic slippages corresponding to different tire-terrain attributes are discussed and analyzed. The impact of the wheel normal reaction on terrain mobility is also considered. Finally, the efficacy of the above methods are evaluated through numerical simulations. Specifically, the terrain attributes and height profile are considered to be stochastic and uncertain while the terrain experiences a drastic change through the simulation. The results prove the efficacy and agility of the control system design.

PI: Dr. Gregory Mocko (Clemson University)
Link to project

Abstract: The development of next-generation ground vehicle systems relies on modeling and simulation to predict vehicle performance and conduct trade studies in the design and acquisition process. In this research, we describe the development of an ontology suite to support modeling and simulation of next generation military ground vehicles. The ontology suite is intended to address model reuse challenges and increase the shared understanding of ground vehicle system simulations. The ontology suite consists of four domain ontologies: Vehicle operations (VehOps), Operational environment (Env), Ground vehicle architecture (VehArch), and Simulation model ontology (SimMod) and an integration ontology. The separate domain ontologies allow for flexibility and extensibility, while the integration ontology establishes semantic relationships across the domains ontologies. The ontology suite, developed using the Web Ontology Language (OWL), leverages existing modeling and simulation automotive standards, including SAE J2998 and SAE J3049, and leverage existing ontologies that include the Basic Formal Ontology (BFO) and Common Core Ontologies (CCO) to ensure semantic compatibility. Based on our experience, we provide recommendations for ontology development and demonstrate the use of the ontology suite to support ground vehicle development. Examples are drawn from the Clemson University Deep Orange development project. The ontology suite is used for modeling the key performance requirements, representing testing procedures to verify the vehicle performance, specifying the vehicle architecture including major systems and subsystems, and capturing model components and simulations. Finally, we discuss how the ontology is used to realize a library of simulation models and identify approaches to support the reuse of simulation models within the vehicle development process.

Jordan A. Whitson, Vladimir Vantsevich, Lee Moradi (University of Alabama, Birmingham); David Gorsich, Brian Butrico, Oleg Sapunkov, Michael Letherwood (GVSC)
Link to project

Abstract: A comprehensive Vehicle Database (VD) was expanded utilizing reputable validated sources. The database is considered live to continually track hundreds of vehicles that fall under the Conventional Armed Forces Treaty (CFET), and a collection of an ever-growing expansion of tracked and wheeled, manned and unmanned vehicles. The VD provides a detailed collection of general vehicle parameters and sub-system technical parameters that can be used for on-demand computing and evaluation of vehicle characteristics required for vehicle modeling and simulation. The VD provides a catalogue for inspection of past and current vehicles and hosts the development of new vehicle conceptual design through parameter analysis.
To assess the technical perfection of new vehicle conceptual design, the Morphing Iron Triangle Profile (MITP) is introduced in the project. Three vertices of the triangle profile are denoted as vehicle properties of terrain mobility, maneuverability, and energy efficiency. The MITP establishes top-possible potentials for each operational property by expanding one vertex for the cost of two others if needed and, thus, morphing the triangle (e.g., enhancing one property for the cost of another).
As analysis shown, for military applications, the generic term of “maneuver” is a tactical concept that is used for planning missions and responses at various levels of military operations. Vehicle complex modeling requires technical integration and new vehicle-level analysis, specifically for decision-making capability of autonomous vehicles to properly manage maneuvering while navigating through severe terrain and operational conditions and responding to dynamic and static obstacles in real time. With this approach, autonomous vehicle maneuverability was defined as a vehicle operational property to characterize the base elements for vehicle performing maneuvers through vehicle turnability, stability, and handling. These base elements of vehicle maneuver were introduced as situational movements (SMs) formed by kinematic patterns that solve the immediate need of navigation in dynamic environment. The SMs were proposed to model the moments of clarity that autonomous vehicles need to perform in combinations to achieve maneuvers.

Vladimir Vantsevich, Masood Ghasemi, Jesse Paldan, Andres Morales (University of Alabama, Birmingham); Michael Cole, Amandeep Singh, David Gorsich (GVSC)
Link to project

Abstract: An electric vehicle with individual wheel drive and steering is an over-actuated system and has a capability of exhibiting agile maneuvering and mobility performance that is unachievable by vehicles with conventional driveline system. Nonetheless, the over-actuation brings new challenges for how to distribute power between wheels and how to steer the wheels. In this presentation, the above problem is addressed for an electric vehicle with individual wheel drives that are powered by AC permanent magnet synchronous motors (PMSMs), and specifically, a hierarchy control and optimization algorithm is developed. A higher level controller provides control actions to ensure trajectory tracking. The controller is based on sliding mode control techniques and utilizes higher order sliding mode disturbance observers and differentiators. The control actions are reproduced by summing up the circumferential forces on all wheels, which are designed by a mobility optimization. Essentially, the circumferential forces and steering angle are designed such that the vehicle performance in terms of mobility is maximized. Further, lower level controls reproduce such forces at each wheel by controlling their torque actions. The lower level controllers are also based on sliding mode control techniques and utilize higher order sliding mode disturbance observers and differentiators to estimate the actual circumferential forces associated with stochastic tire-terrain conditions. Finally, some simulation results and comparisons are provided to demonstrate the efficacy of the algorithm.

PI: Dr. Paul Green, Research Professor of Industrial and Operations Engineering, University of Michigan
Link to project

Abstract: For many driving tasks, there have not been standard, well-defined measures of driving performance, so studies cannot be compared. For example, a basic measure of driving is the longitudinal spacing of vehicles, referred to as headway. Because that measure was the basis of determining traffic flow, the initial reference was front bumper to front bumper of successive vehicles. Now it often refers to front bumper to rear bumper because of emphasis on crash avoidance. For a tractor-trailer, the difference in the definitions is about 55 feet. To help overcome this problem, the author wrote SAE Recommended Practice J2944, Operational Definitions of Driving Performance Measures and Statistics. That document is extensive (170+ pages, more than 300 references).
That document is being revised in the format of a military standard and is being completely rewritten and expanded in its scope. For example: 1) The original standard was for 4-wheel vehicles, for on-road use only. The scope is being expanded to cover other wheel configurations and to include tracked vehicles. Off-road driving performance measures and statistics are being added, such as those for formation driving. 2) The original document did not consider the goals of driving (don’t crash, stay on the road, drive economically, etc.) and how those goals were connected to the measures and statistics of interest. For military combat vehicles, there are additional goals (e.g., being in position to protect other vehicles). 3) There was no connection between models of driving and driving performance measures. 4) Since J2944 was published, considerable research has been conducted, and data from those publications (possibly more than 100) related to currently defined measures need to be added.
The goal of this presentation is to describe provide additional examples of inconsistently defined performance measures and statistics, describe the organization of this new standard, identify some to the measures and statistics being defined, and provide comments on the content of individual definitions.

U of M Investigators: Dr. Alex Gorodetsky (presenter) and Dr. Shravan Veerapaneni; Students: Brian Chen (Mathematics) and Doruk Aksoy (Aero); GVSC: Dr. David Gorsich
Link to project

We propose a computationally faster and more data efficient approach to solving the video-to-action learning problem arising in imitation learning. Our aim is to rapidly learn a player's gameplaying strategy, or soldier behavior in gamified environments, based on video data paired with actions. A common approach to this problem is to first learn a latent space for the video gameplay and then to build action predictors based on this latent space. Deep-learning approaches for identifying this latent space can suffer from large data and computational requirements to learn the latent-mapping. This difficulty is particularly acute in imitation learning applications where limited examples may exist. Instead, we propose a low-rank tensor approach for this video-to-latent space mapping. We discuss recent advances we have made in employing tensor-networks for learning generally, and we then apply it to the imitation learning problem. We show that it outperforms more standard deep learning-based autoencoder approaches in the low-data regime in both video reconstruction and action prediction accuracy. Moreover, we show that the approach yields almost an order of magnitude reduction in computational requirements. These benefits are achieved because the tensor-compression approach does not require extensive architectural and hyperparameter tuning that is needed for deep-learning approaches.

Samuel Misko, Siyuan Zhang, Andres Morales, Masood Ghasemi, Michael Brascome (University of Alabama, Birmingham); David Gorsich, John Brabbs, Paramsothy Jayakumar (GVSC)
Link to project

Accurately simulating off-road vehicle performance in geo-specific unstructured environments relies heavily on accurately modeling the real-world terrain conditions. In this study, methods with a complete toolchain for modeling real-world stochastic terrain conditions (focusing on the stochastic terrain topography and stochastic terrain mechanical characteristics) will be presented. For the terrain topography, Geographic Information Systems (GIS) data from public sources and Ordinary Kriging (OK) method are utilized to generate smooth terrain surfaces at a high-resolution level (up to 5 centimeters). Fractal dimensionality is employed to add stochastic roughness to this smooth terrain to generate the complete stochastic terrain topography. The terrain mechanical properties used for modeling the tire-soil interactions in real-time simulation are defined using Bekker-Wong parameters. The stochastic values for each B-W parameter are defined with probability distribution functions by re-sampling raw bevameter measurement results, and then utilizing Monte Carlo techniques to generate specific values for calculation for each simulation time step. Real-time multi-body dynamics simulations using Vortex Studio demonstrate improved realism for evaluating vehicle performance and mobility during off-road terrain traversal. The data sources, processes, and software (including USGS database, ArcGIS, Python, Meshlab, Vortex Studio, and Blender) are presented for the complete description of the developed toolchain and results.