Technical Talk Abstracts

Day 1: Wednesday June 5

Session 1.A

Project 1.37
Ultrasound perception in physical environments from synthetically trained neural networks
Bogdan-Ioan Popa (PI), Bogdan I. Epureanu (Co-PI), Ganesh U. Patil (Postdoc), Hyung-Suk Kwon (GSRA) (U. of Michigan); Paramsothy Jayakumar (GVSC); Paul Mohan (ZF)

Biosonars utilized by bats and marine mammals excels in environmental perception, surpassing engineered sonars in effectiveness, compactness, and adaptability. Artificial systems approaching the perceptual capabilities of biosonars hold the key to significant advances in imaging and ground vehicle navigation technologies. This talk will present a novel framework for employing Convolutional Neural Networks (CNNs) to decode ultrasound scattering and perceive vehicle surroundings. Specifically, we demonstrate that CNNs trained solely on synthetic data can effectively analyze real echoes to classify objects of different shapes. The innovation lies in careful data augmentation and implementation of multiple parallel CNNs, each specialized to recognize a single shape. The results illustrate the efficacy of the model in differentiating measured echoes that appear perceptually similar. Moreover, an analysis of trained models unveils insights into the diverse acoustic signatures crucial for classification and their susceptibility to perturbations. This synthetic training paradigm for real-world perception obviates the need for laborious and expensive data collection, with potential applications in assisted vehicle navigation.

Project 1.39
Adaptive and Efficient Perception for Autonomous Ground Vehicles Operating in Highly Stochastic Environments under Sensing Uncertainties
Mohammad Al Faruque (PI), Pramod Khargonekar (Faculty), Junyao Wang, Luke Chen, Trier Mortlock (GSRAs) (UCI); Jonathon Smereka (GVSC)

How can adaptive sensor fusion algorithms enhance the perception performance of an autonomous ground vehicle (AGV) deployed with multiple heterogeneous sensors? During AGV operations in challenging, stochastic environments, some sensing modalities negatively impact perception while increasing energy consumption and latency, which can be detrimental to safety-critical missions. Existing perception methods are insufficiently robust in harsh operating environments (e.g., extreme terrain, high speeds, bad weather, low light, sensor obstructions) due to (i) rigidity in their fusion implementations, (ii) their inability to model contextual information in real-time, and (iii) the high levels of sensor input uncertainties present. In this talk, we describe our latest efforts to design adaptive sensor fusion methods that can overcome these challenges by leveraging state-of-the-art computer vision models and ensemble learning techniques. We also present an efficient, uncertainty-aware feature fusion method that can effectively increase perception performance when sensor data is out-of-distribution.

Project 1.40
Touch-Based Sensing for Evaluating Vegetation in Complex Navigation Environments
Christopher Goodin (PI, Mississippi State U.)

An off-road AGV must be able to discriminate between vegetation that is safe to override and vegetation that is unsafe. Prior research on vegetation override of larger trees cannot predict the forces required to override smaller vegetation like shrubs and grass. In this project, we have developed a novel, touch-based sensing modality that directly measures the forces on a vehicle overriding vegetation. Using the measurements from our custom-built, touch-based force sensor, we create a predictor to learn the relationship between observable features (lidar, camera) and physical resistance to traversal. To this end, we present a prediction framework, automatic data labeling pipeline, and initial results for vegetation override force prediction using touch-based sensing. We train our system in a self-supervised manner, retrospectively labeling pre-collision sensor data with the measured override force. Multiple feature extraction, prediction, and training strategies are presented, and their utility and performance are compared.

Project 1.A100
Autonomous Perception for Off-road Terrain Navigation and Mobile Agent Tracking Using Reservoir Computing Techniques
Dr. Mohammed Haider (PI), Samuel Misko, Dr. David Hoxie, Dr. Steven Gardner, Nicholas Bowen, Andres Morales, Dr. Madhurima Maddela, James Michael Brascome (UAB); Paramsothy Jayakumar, Jon Smereka, Ryan Kreiter (GVSC); Arnold Free (Traxara Robotics）

Autonomous vehicle perception has mainly targeted urban areas, lacking diverse datasets for off-road scenes. This project addresses this gap by leveraging efficient reservoir computing concepts for both semantic segmentation of off-road terrain and mobile agent tracking, improving autonomous perception capabilities in military relevant environments. The two perception systems, which integrate Echo State Networks, Graph Neural Networks, Autoencoders, and Convolutional Neural Networks, provide unique performance characteristics while using limited publicly available datasets. The terrain perception system utilizes sensor fusion to generate probabilistic cost-maps for autonomous planners to more successfully navigate unstructured off-road terrain with mobility challenges and variable terrain trafficability. The mobile agent tracking system utilizes minimal layers for clear understanding, improved efficiency, and reduced deployment risk compared to more traditional approaches reliant on complex architectures and other black-box methods. Graph networks based on reservoir computing, along with reinforcement learning, track moving targets over a superpixel field. These approaches demonstrate effective perception enhancements in off-road environments, crucial for robust autonomous systems in military applications.

Project 1.41
Resilient Trajectory Planning for Extreme Mobility on Challenging Slopes
Tulga Ersal (PI), Bogdan Epureanu (Co-PI), James Baxter (GSRA) (U. of Michigan); Paramsothy Jayakumar (GVSC); Chenyu Yi (Mercedes-Benz); Andrew Kwas, Timothy Morris (Northrop Grumman)

This talk presents a novel local trajectory planner, capable of controlling an autonomous off-road vehicle on steep, rugged terrain. Autonomous vehicles are currently unable to operate on steep off-road slopes. The steepness of the terrain necessitates high speeds, yet the roughness of the terrain makes such operation dangerous. Successful navigation requires pushing vehicles to their dynamic limits, which necessitates a complex coordination of control inputs. This project addresses this challenge by using a novel model predictive control (MPC) formulation as the local trajectory planner. A new dynamical model for off-road vehicles on steep and rough terrain is used as the prediction model in MPC. Aggressive operation, including tire liftoff without rollover, is allowed through a new safety constraint. Real-time feasibility is achieved through parallelized GPU computation. A prototype of the planner is presented, and its ability to provide safe, dynamically feasible trajectories is studied through simulation. The results show that the new planner achieves higher success rates than a state-of-the-art baseline, and the relative improvement increases as the vehicle is pushed closer to its mobility limit.

Project 1.A90
Efficient Surrogate Modeling for Reliability-Based Global Path Planning of Off-Road Autonomous Ground Vehicles under Uncertainty
Zhen Hu (PI, U. of Michigan-Dearborn); Zissimos Mourelatos (PI, Oakland U.); Amandeep Singh, David Gorsich (GVSC)

Data-driven surrogate modeling, is essential for accelerating mobility prediction of off-road autonomous ground vehicles (AGVs). This is especially valuable when multi-physics simulations of vehicle mobility are computationally very expensive and uncertainty sources in mobility prediction must be considered. The most commonly used approach to construct a data-driven mobility surrogate model for AGVs adopts a two-stage framework. The first stage focuses on building a cheap-to-evaluate surrogate model using machine learning techniques. The constructed surrogate model is subsequently used for AGV mission planning considering mobility reliability constraints. The separation of surrogate modeling from mission planning in the two-stage approach, however, may require a large number of simulations to train an accurate surrogate model and thus is not practically applicable. This research presents an innovative surrogate modeling method for AGV mobility prediction by coupling uncertainty quantification with AGV mission planning. Results of a case study demonstrate how such coupling drastically reduces the required computational cost for reliability-based global path planning of AGVs under uncertainty.

Session 1.B

Project 3.18
Materials Design of Polycarbonates at the Atomistic Scale with Machine Learning
Christopher Barrett (PI), Doyl Dickel, Micah Nichols (Mississippi State U.)

Neural networks are incredibly useful tools for designing interatomic forcefield models. Using first principles results calculated using density functional theory, large training databases for different configurations of organic molecules can be used to generate fast and highly optimized potentials. However, as is often the case with machine learning, extrapolation beyond the existing dataset is challenging. This is especially true for materials having three or more elements as the possible perturbations in phase space increase. The challenge is mitigated by building known physics relations directly into the neural network formalism so extrapolation should have the correct trends. We have done this by using known physics in the construction of the structural fingerprint which feed into the network and introducing an equation of state which acts as a bounding function for the network output. The neural network result is then treated as a perturbation, reducing errors from the training set. Using this approach, we have built a model for hydrocarbon to serve as a base for a polycarbonate model and demonstrate its stability and reliability.

Project 3.19
Intelligent ultrasound to adaptively control interfacial properties and reactions
Wei Lu (PI), Bogdan Epureanu Bogdan Popa (Co-PIs), Max Nyffenegger, Ganghyeok Im (GSRAs) (U. of Michigan); Katie Sebeck, Matt Castanier (GVSC); Wayne Cai (GM)

Operating hybrid and full electric autonomous vehicles in stochastic and uncontrolled environments puts significant demands on their structure and power system. We aim to develop an ultrasound-enabled adaptive interface control to actively and adaptively change interfacial properties to enable the power system with enhanced adaptivity, readiness, large instant power, and high output capacity. We designed and implemented a setup to deliver ultrasonic waves to a battery cell while measuring its electrochemical performance at the same time. By generating microjets, the ultrasound thins the solid electrolyte interphase (SEI) and enhances ion transport. We showed that ultrasound effectively reduced battery internal resistance, improved usable capacity and promoted fast charging. Modeling and simulations were developed to investigate the effect of ultrasonic waves, which revealed the relation between SEI thinning and cell characteristics. Based on these results, we developed an ultrasound-enabled adaptive protocol for fast charging of lithium ion batteries. The overall charging time is reduced significantly compared to the conventional constant current constant voltage protocol. We further employed metasurfaces for targeted ultrasound delivery and developed a machine learning approach for designing.

Project 3.20
Modeling of a Ground Vehicle Operating in Shallow Water
Ezequiel Martin, Casey Harwood, Hiroyuki Sugiyama (PIs, U. of Iowa)

Accurate prediction of ground vehicle mobility in shallow water, including river crossings, is critically important for reliable mission planning and autonomous navigation. In this study, we develop a fast simulation tool for predicting vehicle mobility in shallow water using a data-driven hydrodynamics model and propose model-scale experiments for validation. To this end, a high-fidelity coupled CFD-MBD model is developed to characterize the hydrodynamic loads exerted on a vehicle in shallow water and used to generate training datasets for the data-driven model. To account for the history-dependent hydrodynamic behavior, LSTM neural networks are introduced, and they are used to predict hydrodynamic loads online exerted on vehicle components in MBD mobility simulations, considering deformable terrain. It is demonstrated by numerical examples that the complex vehicle-water interaction behavior is accurately predicted by the proposed LSTM-MBD vehicle-water interaction model while achieving a substantial computational speedup. Furthermore, ongoing efforts to validate the proposed CFD-MBD and LSTM-MBD models against experimental data will be presented.

Project 3.21
Design of Modular Origami Structures for Multifunctional Cloaking and Protection
Evgueni Filipov (PI, U. of Michigan)

This talk will discuss our progress on exploring deployable thin sheet origami-inspired structures for applications in multifunctional cloaking and protection of ground vehicle systems. Following a broad overview of possible applications, the talk will focus on three connected efforts including lightweight load-bearing origami at the meter scale, deployable acoustic cloaks, and interpretable design for multifunctional origami. We first discuss fundamental principles of origami-inspired systems, and show how the geometric mechanics can be tailored to create thin sheet reconfigurable structures with unusually high stiffness-to-weigh ratios. A physical four-meter-long prototype is demonstrated for its rapid assembly, versatile modular design, and its ability to support structural loads. Next, we present a compactly stow-able system that when deployed creates a structure of thin sheets spaced at prescribed distances. Multi-physical simulations of this structure show that it has anisotropic density and bulk modulus properties and thus has a potential for use in acoustic cloaking. Finally, the talk presents a machine learning approach for inverse design of origami systems. This approach considers categorical and continuous origami features to design for both shape and function of different origami-inspired engineering systems.

Project 3.22
Tire-mud interactions; experimental and numerical studies
Corina Sandu, Alba Yerro-Colom (PIs), Varsha Swamy, Rashna Pandit, Destiny Mason (GSRAs) (Virginia Tech)

As autonomy begins to be integrated into combat vehicles, the necessity for high-fidelity virtual proving grounds becomes imperative. The primary objective of this project is to explore the physics-based modeling of cohesive soil with water content, a crucial soil factor influencing vehicle performance. The worst-case scenario of fully saturated soil in plastic state is studied in detail. In the first half of the talk, we introduce the concept of effective-stress framework in soil modeling for mobility studies. Results from advanced tire-terrain models are presented. Small-scale tests supporting these models including the characterization of the soil and the soil-tire interface are discussed. In the second half, we present the full-scale testing methodology and preliminary results on the Terramechanics Rig at Virginia Tech. The tire is tested on both rigid ground and saturated clay. To enhance the understanding of the short-term and long-term effects, the tire is driven multiple times on the soil at different time intervals.

Project 3.23
Adaptive Structures with Embedded Autonomy for Advancing Ground Vehicles
Dr. Kon-Well Wang (PI), Yuning Zhang, Minh Nguyễn (GSRA), Dr. Patrick Dorin (Postdoc) (U. of Michigan); Dr. Matt Castanier (GVSC); Dr. Jayanth Kudva (NextGen Aeronautics, Inc.); Dr. Ellen C. Lee (Ford Motor Company)

The rapid advances in autonomous systems, such as automated vehicles, have demanded future adaptive structural and material systems to become even more intelligent. Such a need inspired us to advance from the conventional platform that relies mainly on add-on digital computers to achieve intelligence, to mechano-intelligence that embodies intelligence in the mechanical domain. Although interesting studies have attempted, there is a lack of a systematic foundation for constructing and integrating the various elements of mechano-intelligence, namely perception, learning, and decision-making, with sensory input and execution output for engineering functions. In this study, we lay down this foundation by harnessing the physical computation concept, advancing from mere computing to multifunctional mechano-intelligence. As exemplar testbeds, we constructed mechanically intelligent metastructures to achieve wave adaptation via physical computing, and uncover multiple engineering functions, ranging from adaptive noise and vibration controls, wave logic gates, to phononic communication. This research will pave the path to autonomous structures that would surpass the state of the art, with lower power consumption, more direct interactions, and much better survivability in harsh environment and under cyberattack.

Session 1.C

Project 5.19
Adversarial Scene Generation for Virtual Validation and Testing of Off-Road Autonomous Vehicle Performance
Ram Vasudevan (PI), Bogdan Epureanu (Co-PI), Ted Sender (GSRA) (U. of Michigan); Mark Brudnak (GVSC); Reid Steiger (Ford)

Developing autonomous vehicles (AVs) that operate in diverse and demanding environments is a difficult challenge. Two fundamental tools that can accelerate this process are testing an AV in diverse simulated environments and identifying core system weaknesses. While most efforts focus on improving these tools for on-road AVs, this work focuses on an analogous set of tools for off-road AVs. A method called Black-Box Adversarially Compounding Regret Through Evolution (BACRE) is proposed for identifying adversarial scenarios using an evolutionary algorithm guided by a novel regret-based metric for general navigation tasks. A black-box approach is often preferable when system complexity can be diverse, like with off-road AVs, and when whole-system testing is required. A custom simulation platform is also provided to assist with the automated testing of AVs in diverse, unstructured environments. Numerical experiments demonstrate that BACRE's evolutionary process gradually increases scenario complexity to degrade vehicle performance (an effective and explainable process that comparable methods cannot achieve). Consequently, BACRE can streamline AV development by finding weaknesses at various development stages. We will also discuss BACRE’s limitations and suggest potential future work.

Project 5.21
Brain Inspired Probabilistic Occupancy Grid Mapping with Hyperdimensional Computing
Maryam Parsa (PI), Shay Snyder (George Mason U.); David Gorsich (GVSC); Andrew Capodieci (Neya Robotics)

Real-time robotic systems require advanced perception, computation, and action capability. However, the main bottleneck in current autonomous systems is the trade-off between computational capability and energy efficiency. World modeling, a key objective of many robotic systems, commonly uses occupancy grid mapping (OGM). This approach divides the environment into discrete cells and assigns probability values to attributes such as occupancy. Existing methods fall into two categories: traditional methods and neural methods. Traditional methods rely on dense statistical calculations, while neural methods employ deep learning for probabilistic information processing. By integrating cognitive science and hyperdimensional computing, a theory of neural computation based in hyperdimensional computing has emerged. In this study, we propose a Fourier-based hyperdimensional OGM system, combining the interpretability and stability of traditional methods with the improved computational efficiency of neural methods. Our approach, validated across multiple datasets, achieves similar accuracy to traditional methods while reducing latency by 79.4x and 3.7x. Moreover, we achieve 1.56x latency reductions compared to neural methods while eliminating the need for model pretraining.

Project 5.22
Unsupervised Testing and Verification for Software Systems of Ground Autonomous Vehicles
Nickolas Vlahopoulos (PI), Sean Hickey (GSRA) (U. of Michigan)

The overall objective of this project is developing a new unsupervised software testing approach. The term “unsupervised” indicates expecting minimal human effort in defining test cases and expected outcomes. Work completed during the first seven months of this effort is presented. Two simulation harnesses were established at UM. Both use the ROS2 similar to the Robotics Technology Kernel (RTK). The simulation harnesses are used for developing and demonstrating the new capabilities before transferring them to RTK. The public domain Autoware Universe system for operating autonomous vehicles comprises a comprehensive simulation harness comparable to RTK. A second, much simpler simulation harness, was also established in order to implement and demonstrate the new developments in a much faster turnaround time. An interceptor code and a code for monitoring the error metric evaluation are developed and placed within the software system which is tested. The former intercepts the data flow, alters it and publishes the altered data flow. The monitoring code identifies successful termination, inactivity, excessive run time and unexpected termination. The information from the monitoring code will be used for developing an error metric. A capability that can execute multiple runs for the simulation harness and for multiple generations has also been developed.

Project 5.23
Automated Co-Design of Vehicles and their Teaming Operations for Optimal Off-Road Performance
Bogdan Epureanu (PI, U. Michigan)

The engineering process of AI-powered autonomy involves conceptualization, designer input, intelligence empowerment, and performance experimentation. This process becomes especially challenging when developing multi-agent systems such as human autonomy teaming due to complex interactions between agents and operation environments. Existing approaches either solve teaming strategies given pre-defined physical attributes of vehicles or find attributes that satisfy pre-defined strategies. In this work, we devise a more general approach to simultaneously co-design team physical attributes and teaming strategies. We present methods using an iterative heuristic-based process and a genetic algorithm that perform co-evolution of design attributes and team behavior. Furthermore, we propose a decision transformer architecture that leverages a sequence modeling and attention mechanism to draw dependencies between physical vehicle attributes and behaviors. We demonstrate the co-design methods in a multi-agent logistic operation considering vehicle delivery capability, route traversability, adversaries, and constraints on physical attributes. This effort aims to automate the co-design process that cost-effectively leverages available resources and assets for maximum teaming effectiveness, and it also fast adapts to the expectations and changes in operation environments.

Project 1.A73
Quantum Computing Innovation for Off-Road Mobility
Shravan Veerapaneni (PI), James Stokes (Faculty), Sam Cochran, Oliver Knitter, Rohan Kodati (GSRAs) (U. of Michigan); Jeremy Mange, Paramsothy Jayakumar, David Gorsich (GVSC); Saibal De (Sandia National Lab)

Variational quantum algorithms (VQAs) and noise-robust variants are emerging as promising approaches to leverage quantum and quantum-inspired computing capabilities, particularly in the face of existing hardware limitations. In this talk, we demonstrate the practical applications of VQAs to two important problems arising in the field of autonomous ground vehicle systems research. Firstly, we demonstrate the effectiveness of Variational Quantum Linear Solver, and its classical counterpart, within a minimum-map Newton solver framework to physics-based offroad mobility simulations. By leveraging these algorithms, we demonstrate how to map the computationally-intensive part of the complementarity problem, modeling the frictional contact, onto to a near-term quantum device, thereby, paving the way to exploit the quantum advantage in overcoming the curse-of-dimensionality induced by the multi-scale nature of the terramechanics in offroad mobility. Secondly, we showcase an improved version of the Quantum Approximate Optimization Algorithm (QAOA), a noise-robust VQA, and its application to clustering problems in autonomous mobility. We introduce a method to enhance QAOA accuracy by warm-starting the initial quantum state with solutions from Max-Cut relaxation and illustrate its performance on currently available quantum devices.

Project 5.A104
Real-Time Optimization of Autonomous Ground Vehicle Simulations: A Parallel Co-Simulation Framework Approach
Dr. Vladimir Vantsevich (PI), Sam Misko, Dr. David Hoxie, Nicholas Bowen, Andres Morales, James Michael Brascome, Pi Viriyaphap (UAB); Dr. Paramsothy Jayakumar (GVSC); Arnold Free (Traxara Robotics); Torsten Kluge (dSPACE GmbH)

Critical to the improvement of autonomous ground vehicles for military-relevant scenarios, is the ability to simulate the dynamics, mobility, sensor measurements and more while operating in real-time. Simulations are often compartmentalized to allow for modularity and/or the use of multiple software packages, which necessitates the development of equally sophisticated co-simulation frameworks. These simulations and models can require different operating frequencies and compute resources making it difficult to realize a robust synchronous platform running in real time. Many current optimization approaches lack dynamic load balancing and/or fail to properly utilize distributed computation. We propose a sandbox solution for parallel co-simulations, optimizing computational frequency with a real-time controller. Our results show performance enhancement while synchronizing simulations to wall time. Our research acknowledges the trade-off between computational time and model fidelity, in which we develop mathematical methods to evaluate vehicle and sensor model fidelity. This research provides techniques to monitor real-time co-simulation frameworks, enabling closed-loop interactions among disparate software applications. This research provides a co-simulation framework solution that can be used to achieve optimal fidelity in real-time for simulating vehicles in unstructured terrain.

Day 2: Thursday June 6

Session 2.A

Project 2.15
In-the-wild Question Answering: Toward Natural Human-Autonomy Interaction
Rada Mihalcea (PI, U. of Michigan)

Current autonomous vehicles efficiently explore vast, uncharted terrains but struggle to “report back” the information they collected in a manner that is easily accessible to human users and does not produce information overload. Our focus is on in-the-wild multimodal question answering, in which given a video recording of an in-the-wild territory (as obtained by e.g., a vehicle charting a new territory), we can answer questions about the entities, things, and events it has observed (e.g., “What objects do you see?”, “What is blocking the road?”), while also providing visual support for the answers. In particular, given that a major shortcoming of current multimodal models is their inability to represent compositional relations between entities in the data, we introduce a framework to significantly improve the ability of existing models to encode compositional language, with over 10% absolute improvement on compositionality benchmarks, while maintaining or improving the performance on standard object-recognition and retrieval benchmarks.

Project 2.16
Language Communication and Collaboration with Autonomous Vehicles Under Unexpected Situations
Joyce Chai (PI), Yidong Huang, Jacob Sansom, Ziqiao Ma (U. of Michigan); Felix Gervits (ARL)； Chris Mikulski (GVSC)

Recent advancements in foundational models (FMs) have unlocked new prospects in autonomous driving, yet the experimental settings of these studies are preliminary and over-simplified, which fail to capture the complexity of real-world driving scenarios in human environments. It remains under-explored whether FM agents can handle longer-horizon tasks with free-from dialogue, and deal with the unexpected situations caused by sensor limitations, environmental dynamics, or plan changes. To explore the capabilities and boundaries of FMs faced with the challenges above, we introduce DriVLMe, a video-language-model-based agent to facilitate natural and effective communication between humans and autonomous vehicles that perceive the environment and navigate. Developed and evaluated in CARLA simulator and human dialogues, DriVLMe surpasses existing baselines in open-loop assessments using the Situated Dialogue Navigation (SDN) benchmark and exhibits promising navigation and replanning abilities in closed-loop CARLA simulations.

Project 2.18
Investigating Required Transparency Information and Display Features through an Empirical Study using a Dual-tasks HAT Simulation
Sang-Hwan Kim (PI), Dasol Han, Seungju Choi (GSRAs) (U. of Michigan-Dearborn); Kayla Riegner, Jonathon Smereka, Terrance Tierney (GVSC); Yu Zhang (Denso)

In Human-AI Teaming (HAT), the enhancement of situational awareness and trust plays a pivotal role in elevating task performance. Achieving this enhancement involves the effective provision of transparency, elucidating the actions and rationale of the AI system with effective communication between human operators and AI. The study is to explore the features of HAT displays that offer optimal transparency information, specifically in dynamic, multitasking situations such as battle, where AI supports human operators.
This presentation introduces the outcomes of the first-year study, including the development of a dynamic dual-task simulation platform and the results of the initial experiment aiming to discern natural communication methods among human operators collaborating on simulation tasks. Ten Pairs of participants assumed the roles of main operators and virtual AI assistants, engaging in the experiment. Types of collaboration strategies, behaviors, and communication forms based on natural interactions were identified through the experiment. The observed behavioral patterns and communication methods will be integrated into the AI and communication formats in the second-year study, contributing to the determination of optimal transparency features in the task scenarios.

Project 2.A101
Communication, Reaction, & Effectiveness in Warfare (CREW)
Despina Stavrinos, PhD, Karlene Ball, PhD, Benjamin McManus, PhD, Piyush Pawar, Andrea Underhill, PhD, Samuel Misko, James Michael Brascome (UAB & UA); Victor Paul, Terry Tierney (GVSC); Brandon Perelman, PhD, Gregory Gremillion PhD (ARL); Thomas Anthony (Analytical AI, LLC)

Advancing technology provides enormous potential for tasks previously completed by humans to be managed by autonomous systems. The optimal quantity and delivery method of assistance, currently unknown, is critical for successful artificial intelligence (AI) integration into military vehicles. The overarching objective of this project is to develop an ecologically valid, high-fidelity experimental simulation to assess crew performance and communication under varying levels of AI assistance. The chief contribution of the proposed project will be to evaluate the impact on mission efficacy when automated assistance is introduced to a 6-member crew. Analytical prediction models will be developed to determine (1) the ideal number of automated assistance technologies used, and (2) the minimal real-world efficacy of automated assistance required to benefit mission objectives. Models quantifying the amount of cognitive resources freed by the automated assistance will be developed based on the benefits of the assistance to (1) situational awareness, (2) driving performance, and (3) team communication.

Project 1.A81
Mathematical Approaches for Learning From Gaming Data
Alex Gorodetsky, Shravan Veerapaneni (PIs), Doruk Aksoy, Brian Chen (GSRAs) (U. of Michigan); David Gorsich, Joseph O'Bruba (GVSC)

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 game playing strategy, or soldier behavior in gamified environments, based on video data paired with actions. In previous work, we have demonstrated the effectiveness of using incrementally trained tensor networks to extract latent representations from data, which can then be used as input for behavioral cloning. In particular, the incrementally trained tensor network yields almost an order of magnitude reduction in computational requirements compared to neural network approaches like autoencoders and variational autoencoders. However, existing tensor network approaches can fall short at first-person-perspective games with more complex three dimensional visuals such as Minecraft. Due to these limitations, we developed an incremental Hierarchical Tucker decomposition algorithm, which will be the first in the literature. We find that this tensor network method leads to improved compression times compared to our previous work using an incremental tensor train, allowing us to tackle more complicated games. We present preliminary results of this approach, including visual reconstructions and using the extracted features for behavioral cloning on the MineRL competition dataset.

Project 1.A113
Responsible AI-Based Control of Unmanned Ground Vehicles in Severe Dynamic Terrain Environments
Masood Ghasemi (PI), Vladimir Vantsevich, Lee Moradi (Co-PIs), Hunter Zhang, Noah Wolf (WPI); David Gorsich, Amandeep Singh, Michael Cole (GVSC), Jianbo Lu (Nkkola Corp.)

Analytical approaches to maneuver control of vehicles mainly rely on well-defined scenarios and conservative problem definitions. Thus, they may face limitations on their application in some realistic scenarios especially with severe dynamic terrain and environmental conditions. Adaptation and learning capabilities are key features of soft-computing or AI-based control approaches that can address the above challenges. Nonetheless, several aspects of the AI technology including augmentation or replacement of human judgment, safety challenges, data and algorithmic bias, etc. pose ethical and trust issues. In response to the above concerns, the DoD has adopted some AI ethical principles to ensure responsible behavior, processes, and outcomes in systems involving AIs. This project researches the design of an AI-based maneuver control of a vehicle with distinct optimal features for mobility, energy efficiency, and maneuverability. Furthermore, it researches the design and implementation of a responsible AI (RAI) framework based on the DoD ethical principles to ensure trust in the system. In this presentation, the initial results on the RAI framework design comprising the AI-based control architecture, AI data generation to mitigate bias and unintended behaviors are discussed.

Session 2.B

Project 3.24
Additively Manufactured All-metallic Metamaterial Solutions for Protection of Electronic Systems in Autonomous Vehicles
Lorenzo Valdevit (PI), Diran Apelian (Co-PI) (UCI)

Abstract TBA.

Project 3.A112
Advanced Manufacturing of composites using Robotic fiber placement, Novel Multi-material Joining, and Integrated Sensors
Mahmood Haq (PI, Michigan State U.)

Abstract TBA.

Project 3.A111
Multi-Band Communication Antenna Systems for Ground Vehicles Enabled by Advanced RF Packaging Integration
John Papapolymerou (PI, Michigan State U.)

Abstract TBA.

Project 3.A103
Evaluation of Under-body Blast Response to Loading from Alternative Terra-Medium Environments
David Littlefield (PI, UAB); Kumar Kulkarni, Miriam Figueroa-Santos, Jacob Woten (GVSC)

In this research project we have implemented coupling of proven Lagrangian/ALE methods with the widely used Discrete Element particle method, to improve computational models with hard, coarse aggregates in addition to the softer media. In this approach, aggregates (rocks) are modeled using discrete elements and softer media (soil) using Lagrangian/ALE elements. Example calculations are shown where a buried C4 charge is loading a rolled homogeneous armor (RHA) plate. Results illustrate that the Lagrangian particles (rocks) play a significant role in the localized deformation of the plate. This has important consequences for the design of protection strategies in armored vehicles.

Project 3.A106
Novel materials for water purification to ensure water availability in the combat zone
Anja Mueller (PI), Bradley Fahlman, Itzel Marquez (Co-PIs), Emmanuella Anang, Catriana Nicholls, Abolade Busari, Gabrielle Johnson (Central Michigan U.); James Dusenbury (GVSC)

Military personnel encounter water contamination issues when deployed in locations where clean water is not readily accessible. There are health risks associated with consuming water containing heavy metals and other contaminants like arsenic and ammonia. Water treatment techniques including adsorption and membrane filtration can be used to remove arsenic and ammonia from water, but these commercial adsorbents and membranes lack high selectivity for the contaminants. In this study, selective adsorbents and membranes were developed through imprinting polymerization. The properties of the materials were enhanced by increasing the contaminant template amount in the polymer and incorporating graphitic carbon nitride. The adsorption capacity of the materials improved with the addition of graphitic carbon nitride. The underlying adsorption mechanism has been studied. Further optimization of the material and research on the influence of other environmental factors are required to ensure the safety of military personnel on the battlefield.

Session 2.C

Project 1.A107
Integrated Design and Efficient Safe Control for Terrain-Adaptive Ultra-lightweight Vehicles
Zhaojian Li (PI, Michigan State U.)

Abstract TBA.

Project 1.A108
Self-Powered Wireless Sensing Platform for Vehicle Attitude Control
Nizar Lajnef (PI, Michigan State U.)

Abstract TBA.

Project 4.37
Risk Averse Vehicle Energy, Thermal Signature Management and Control to Enable Silent Mobility/watch
Jeffrey Naber (PI, Michigan Tech)

Abstract TBA.

Project 4.A109
Develop Lightweight, Low-temperature, and Safe Batteries for Autonomous Electric Vehicles
Chengcheng Fang (PI, Michigan State U.)

The weight of battery packs in an electric vehicle (EV) can reach 25% of the total weight, making EVs 10 - 15% heavier than gasoline-based cars. The heavyweight decreases the EV’s dynamic and braking characteristics and reduces the mileage that batteries can power. Reducing the battery weight, improving its energy density, and enhancing its wide temperature performance are essential to enable viable EVs for Army applications. This project includes multi-dimensional efforts from developing new battery chemistry at single cell level to developing lightweight and swappable battery packs, to achieve high-performance battery packs for EVs. In this presentation, we update the electrolyte systems we have explored to enable high-energy lithium batteries with improved low-temperature performances.

Project 4.A110
Lightweight Electric Powertrain with High-speed Machines and Drives
Shanelle Foster (PI, Michigan State U.)

Abstract TBA.

Project 4.A105
De Novo Design of Energy Storage Materials Through a Synergistic Approach
Bradley D. Fahlman (PI), Veronica Barone, Valeri G. Pekov (Co-PIs), Aliakbar Yazdani, Jyoti Pandey, Mukesh Jakhar, Benjamin R. Seltin (Central Michigan U.); Yi Ding (GVSC)

Metal-sulfur batteries, exemplified by lithium-sulfur (Li-S) batteries, represent a promising frontier in high-energy-density energy storage solutions. Despite their potential, Li-S batteries encounter challenges like the shuttle effect, hindering cyclability and capacity retention over numerous cycles. Addressing these issues demands advanced materials science and engineering approaches for the cathode. Techniques like defect engineering in graphitic carbon nitride and the utilization of single-atom catalysts show promise in overcoming these hurdles. Our project includes the synthesis, characterization, electrochemical testing, and density functional theory (DFT) calculations to understand the interactions between defect sites in the molecular structure of graphitic carbon nitride and transition metal single atoms. These findings have yielded advancements in Li-S battery technology, enhancing charge/discharge rates, and bolstering capacity retention. Additionally, we have examined the structural defects and properties of graphitic carbon nitride, elucidating its benefits toward the anchoring of single atom catalysts.