2021 ARC Research Seminar - Winter Series
January 15, Friday, 9:00-10:30am eastern time
ARC Cluster of Activities at University of Alabama at Birmingham
Dr. Vladimir Vantsevich, Director/PI of AVMI Professor of Mechanical Engineering
Dr. Lee Moradi, General Manager/Co-PI of AVMI Professor of Mechanical Engineering
Dr. Mohmmed Haider, Associate Professor of Electrical Engineering
Dr. Bala Balendran, Lead CAE Scientist at AVMI
Dr. Lauren Reinerman-Jones, Director of Autonomous Mobility Simulation and Training Lab
Samuel R. Misko, Program Coordinator of AVMI
The US Military is undergoing a rapid transformation, in response to changing global security environment and reemerging challenges from adversaries developing new cyber, electronic, and conventional capabilities. This creates a need for rapid modernization of key capabilities, including advanced autonomous systems. In particular, the development of military vehicles focused on radically enhanced mobility where ability to traverse diverse off-road terrain, with varied slopes and elevations under various adverse environmental conditions would enable future U.S. ground forces to tackle varied and unpredictable combat situations. The seven exploratory and translational projects that have begun at UAB as part of the newly established Autonomous Vehicle Mobility Institute (AVMI) development group are focused on research of novel methods, approaches, and enabling technologies to improve the state of the art in a number of areas directly related to autonomous vehicle mobility. Each of these projects were developed such that their work products complement or contribute each other and will together will form the basis of advanced technology demonstrations on AVMI’s Simulator of Autonomous Mobility. These projects will be presented so as to inform new collaborative and synergistic opportunities with other ongoing and/or future ARC efforts.
A Hybrid Controller with Observation And Decision Making for Autonomous Mobility Control System (Thrust Area 1 project)
This research study develops fundamentals for an autonomous mobility control system (AMCS) with three components: a) an adjustable mobility control algorithm that is able to effectively operate in different severe terrains within an appropriate response time that is close to the tire-soil longitudinal relaxation time constant and, thus, to prevent the tire spinning, b) observation algorithms that are able to provide on-line states of the single-wheel module’s normal, longitudinal, and rotational dynamics, and c) an Artificial Intelligence (AI) - based learning component that is able to learn from the observers’ data in real-time and improve/mature the mobility control parameters. The autonomous mobility control system that functionally integrates the above-listed three components will be mathematically designed and demonstrated in computational simulations for a single-wheel module that is a simulation model of a quarter of a 4x4 vehicle with parameters of FED-Alpha including a single wheel-tire system, propulsion system (electric motor and driveline), brake, suspension, and steering.
Unique Approaches to Utilization of Proprioceptive and Exteroceptive Sensor Systems for Autonomous Agile Mobility (Thrust Area 1 project)
This project seeks to develop and define new fundamental approaches and provide recommendations for the utilization of proprioceptive and exteroceptive sensor system for autonomous agile mobility. In contrast to the existing modern electronic vehicle control systems with “reactive control”, the response time in the proposed “agile control” is required to be less than the vehicle’s tire relaxation time. To facilitate the development of this new vehicle mobility paradigm, a similar paradigm shift in wheel speed sensor technology is required to provide feedback to the vehicle’s control system with an unprecedented level of temporal and spatial accuracy. Another challenge for autonomous vehicle mobility is the detection and classification of moving objects, where the data representation plays a critical role for better accuracy and performance. The innovative sensor fusion and in situ (i.e., edge) reservoir computing will enable faster computation, sensor information extraction, and intelligent decision making.
Instant Tire Slippage Characterization with Digital Image Correlation for Autonomous Mobility Applications (Thrust Area 3 project)
The project is aimed at conducting the basic research to uniquely define parameters (‘cause’) which are measurable in real-time and which lead to tire slippage (‘effect’). Specifically, this project belongs to the (i) terra-mechanics area of detecting dynamic changes in the tire-soil interaction, and (ii) in-tire sensing for its use in AI-based algorithm design for autonomous vehicle control. Approach of this project is broken into the following four parts: (1) Creation of databases for traction characteristics, slippage, and instantaneous compression based rolling radius for different tires on different soils, (2) Digital image correlation of strain and deformation fields of tire outer surfaces in agile dynamic tire tests to wheel torque, the normal load and the road velocity, (3) Finite element simulation of agile dynamic tire tests, correlation of FE strain and deformation fields to those of measured in physical test, and (4) Establishment of a functional relationship between strain field on the internal surfaces of tire with slip behavior. Approaches (1) and (2) are continuation efforts of previous ARC project efforts.
Autonomous Vehicle Interactive Dynamics and Morphing with Mobility & Maneuver Self-Learning-and-Improvement (Thrust Area 1 project)
Dynamic couplings in vehicle systems require new approaches to de-couple and establish interactive dynamics of the systems. This project shall study two types of dynamical couplings: couplings of the physical components of vehicle cyber-physical systems and couplings of physical systems and information processes in hybrid and electric vehicles. A method will be designed to decouple dynamical actions of a vehicle's driveline from its steering to complement vehicle mobility and maneuver. The method will be applied to two vehicle configuration cases: a 4x4 FED-Alpha and an autonomous ground vehicle (AGV) composed of a flexible manipulator and rigid body. An optimization problem will be formulated and solved to determine the optimal split of the summation of the circumferential wheel forces among the 4x4's individual wheels in a way that maximize mobility and/or maneuver of the vehicle. For the AGV, the morphing of the conﬁguration by dynamically reconfiguring the manipulator is proposed to manage instant magnitudes of the moments of inertia of the vehicle and, thus, control the tire-ground forces for the purpose of improving navigation, mobility, and maneuver. A system for autonomous mobility & maneuver self-learning and improvement will communicate with the vehicle’s perception, navigation and planning system and an RL-based self-learning algorithm of the mobility and maneuver estimation/observation system to establish safe perimeters when moving along a given trajectory path.
Assessment and Virtualization of Tire-Soft Soil Interactions for Real-Time Evaluation and Control of Autonomous Vehicle Mobility (Thrust Area 3 project)
Current methods for modelling and simulating autonomous vehicle-terrain interactions are insufficient for Soldier training and mission rehearsal. The terramechanics approach is tedious and does not scale efficiently or effectively, leaving a time lag for vehicle response in simulation, which is noticeable to the human operators. Large geo-spatial databases exist for local network housing or updating from cloud servers, but the level of detail required for realistic physics-based interaction with an autonomous vehicle lacks in simulation. This project is addressing those deficits by identifying modeling methods, developing algorithms and requirements, and implementing the best toolchain for deducing autonomous vehicle-terrain interaction in a simulated environment for adequate functional and physical fidelity matched to operator mental models for enhanced training outcomes.
Maximizing Autonomous Mobility and Energy Efficiency On-the-Go Through Exteroceptive and Proprioceptive Self-Learning-and-Improvement (Thrust Area 4 project)
Mechanical driveline systems should be developed that are flexible enough to provide various power splits to the driving wheels for the purpose of either energy efficiency or mobility. In case of individual electric drives, new analytical fundamentals will be developed to coordinate the power delivering to the wheels, which lack mechanical connections in fully electric vehicles. To overcome these challenges, new vehicle dynamics fundamentals shall be developed to mathematically formulate and solve the problem of the wheel power distribution and to establish conditions for max terrain mobility. By correlating the new analytical accomplishments in mobility and energy efficiency management with the distinctive features and requirements to the autonomous vehicle models, research directions are identified and formulated for developing AI-based fundamentals to benefit the autonomous mobility and energy efficiency management. The overall goal of the project is to develop fundamentals and simulate an AI-based system to maximize mobility and energy efficiency of a 4x4 vehicle with e-motors in the wheels.
Technical Approaches and Analysis of Vehicle Conceptual Design for Mobility and Autonomous Mobility (Thrust Area 3 project)
The ultimate purpose of the project is to develop an advanced methodological foundation for vehicle conceptual design and vehicle major subsystem design for mobility and autonomous mobility. The work will begin with an analysis of technical approaches from eastern and western countries and will provide a comprehensive description and analysis of differences and similarities in analytical approaches to defining, modeling and assessing mobility, maneuver, and movability. A special attention will be on intelligent information management for mobility of autonomous ground vehicles (AGV), including approaches to modeling and simulation of AGV sub-systems, path planning and control, and observation and control of tire slippage and mobility. Parameters and characteristics of various types of wheeled and tracked, manned/unmanned vehicles will be gathered from reputable multiple open sources to build a comprehensive database for the purpose of establishing design metrics for mobility and related mobility metrics that are suitable in the eastern and western approaches. An express method will then be developed and applied to assess the eastern and western approaches to vehicle mobility or movability and to range the vehicles quantitatively and qualitatively.
Based on the above-described analysis, a method of conceptual vehicle design for mobility, which involves integrating vehicle dynamics with dynamics of sub-systems and intelligent decision-making and targets to identify the best combination(s) of the main vehicle parameters for mobility, will be developed further in the project. The approach will be based on inferring and analyzing regression models of the main vehicle conceptual parameters, which cover three clusters, including vehicle mass and overall geometry, vehicle powertrain, and chassis.
In collaboration with the NATO AVT-341 group, the integration of past, current, and future frameworks of vehicles and AGV design metrics in this project will provide a NATO STANREC (useful practices) on AGV mobility modeling, simulation and assessment.
These ARC projects support a collaborative Ground Vehicle Alliance around the ARC, which now includes the ARC, the Autonomous Vehicle Mobility Institute (AVMI) at the University of Alabama Birmingham, and the Virtual Prototyping of Ground Systems (VIPR-GS) Center at Clemson University.
January 29, Friday, 9:00-10:30am eastern time
Evaluating Sensitivity of Autonomous Algorithms to Sensor Error and Environmental Conditions
While much effort has been devoted to the development of autonomous navigation algorithms in recent years, techniques and procedures for quantifying the performance of these algorithms have been developed on an ad-hoc, case-by-case basis. In particular, the use of system-level tests to compare the performance of autonomy subsystems leads to a high degree of uncertainty in algorithmic performance. This research project set out to develop a simulation-based framework to support the systematic evaluation of autonomous ground vehicle (AGV) software with metrics at both the system and subsystem level. Within the framework, perception and planning algorithms were tested to explore the sensitivity of algorithm performance to variations in sensor quality, environmental conditions, and operational requirements. Our results showed that subsystem-level metrics were not strongly correlated to system-level performance, indicating that system-level testing is appropriate for most AGV.
Terrain Adaptive Autonomous Vehicles for Uncertain Off-Road Environments
Autonomous ground vehicles (AGVs) are critical for the future of the military. As military vehicles often need to operate with high mobility even in off-road environments with deformable terrains, AGVs need to meet the same requirements. However, current navigation strategies do not satisfy these needs due to one or more of the limiting assumptions such as rigid terrains, linear operating regimes, or low speeds. Furthermore, terrain properties can vary significantly during the operation of the vehicle and cannot be directly measured with the sensors available on-board. If the trajectory planning algorithms that navigate these vehicles are not aware of such changes, they can devise plans that are difficult or even infeasible for the vehicles to execute, leading to performance and safety issues. Therefore, the overarching goal of the project is to develop, implement, and evaluate a terrain-adaptive trajectory planning algorithm for improved safety and performance of autonomous vehicles in off-road conditions.
This presentation highlights current progress in achieving this goal. First, due to limitations in operating conditions, computational efficiency, and continuous differentiability of state-of-the-art terramechanics models, a novel model is presented to satisfy the differentiability requirements of optimal control while also remaining efficient and capable of dynamic operation. Second, state-of-the-art terrain estimation algorithms face limitations in application due to sensor setup, model simplifications, computational efficiency, and requiring low speed vehicle operation. To address these shortcomings, a method is presented to estimate the terrain properties online from measurements available on-board the vehicle, achieving less than 5% estimation error. Third, this presentation highlights the development, and improved safety and performance of a single-level adaptive model-predictive trajectory planning and tracking algorithm that uses the estimated terrain properties and the new terramechanics model to continuously adapt to the environment and improve its model-based mobility predictions while optimizing the trajectory.
March 12, Friday, 9:00-10:30am eastern time
Quantum Computing Innovation for Off-Road Mobility
Recent impressive progress in quantum technology, particularly in programmable quantum computers, has invigorated a renewed interest in quantum algorithm research. This excitement is largely fueled by the existence of quantum algorithms exhibiting exponential speedups compared to the best-known classical algorithms (such as Shor factoring). While current quantum algorithms research is predominantly focused on discrete optimization problems, our goal in this project is primarily on continuous optimization problems arising in off-road mobility simulations. To this end, we developed a framework for continuous-variable neural quantum states (cNQS), which employs a quantum rotor model and neural networks to simulate variational quantum states. In particular, we will discuss a quantum generalization of a classical heuristic algorithm (Burer-Monteiro-Zhang) for solving the MaxCut problem and present results on solving it using cNQS.
Enhanced Multiscale Off‐Road Mobility Prediction Capability with Machine Learning Constitutive Modeling for Large Deformable Granular Terrain
A high-fidelity off-road mobility model is an integral part of the Next-Generation NATO Reference Mobility Model (NG-NRMM) to enable reliable operational planning. Modeling of complex granular terrain dynamics requires considering not only grain-scale particle dynamics in the order of millimeters to micrometers, but also large soil deformation on macro scale which accounts for the physical interaction with tires traveling over a wide range of terrains. In consideration of the multiscale aspect in the tire-soil interaction problem, the hierarchical FE-DE multiscale off-road mobility model was proposed in the previous years, and it has proven to be effective in eliminating limitations of existing single-scale FE and DE terrain models. In the multiscale terrain model, speedup of the lower-scale DE RVE computation is essential to enable quick mobility prediction. To address this issue, this talk presents a new hierarchical multiscale off-road mobility model enhanced by a data-driven neural network surrogate model that can capture the complex material behavior of granular terrains. The predictive ability as well as effective training procedures of the neural network surrogate model are discussed for various off-road mobility simulation scenarios.
April 2, Friday, 9:00-10:30am eastern time
Remote connection via Microsoft Teams. Contact William Lim (email@example.com) for details.
The urgent modernization needs of the U.S. Army require new approaches to enable faster prototyping and experimentation. Virtual prototyping is more time and cost-efficient than physical prototyping, enabling faster fielding of technology and reducing risks by allowing failure at low cost. The main goal of this project is to create autonomous driving tools that can be integrated into a simulation platform providing such capabilities. Our approach is to develop a sensor validation tool that uses a set of performance metrics to validate the fidelity of real and virtual sensors, and a scene recreation tool to create an accurate virtual representation of the real world. To ensure sensor fidelity, we use these tools in combination and follow an iterative process of tuning the virtual sensor models. This also helps us determine the minimum resolution needed to model virtual scene elements such that virtual and real perception performance are nearly identical.
Probability of Mobility for Mission Planning of Autonomous Ground Vehicles at “High Stress” Environments
PIs: Dr. Zissimos Mourelatos (Professor, Mechanical Engineering, Oakland University) and Dr. Zhen Hu (Assistant Professor, Industrial and Manufacturing Systems Engineering, University of Michigan, Dearborn)
Link to project
Vehicle mobility of off-road autonomous ground vehicles (AGVs) is uncertain due to the presence of heterogenous uncertainty sources, such as model uncertainty in modeling and simulation, natural variability in terrain and soil properties, etc. The uncertainty of vehicle mobility poses various challenges to the mission planning of AGVs in complex off-road environment. This project addresses the uncertainty in vehicle mobility based on three main developments, namely (1) design of tests to collect the most informative data for uncertainty reduction of vehicle mobility; (2) model updating of vehicle mobility models; and (3) reliability-based mission planning of off-road AGVs under uncertainty. In this talk, we will first briefly summarize the first two developments. We will then present details of a reliability-based mission planning for off-road AGVs. The developments of this project allow us to improve the prediction confidence of vehicle mobility and also guarantee the reliability of AGVs in uncertain off-road environments.