2020 ARC Research Seminar - Fall Series
September 25, Friday, 9:30-11:00am eastern time
Trust-based Symbolic Motion and Task Planning for Multi-robot Bounding Overwatch
Multi-robot bounding overwatch requires timely coordination of robot team members. Symbolic motion planning (SMP) can provide provably correct solutions for robot motion planning with high-level temporal logic task requirements. This talk will present a framework for safe and reliable SMP of multi-robot systems (MRS) to satisfy complex bounding overwatch tasks constrained by temporal logics. A decentralized SMP framework is first presented, which guarantees both correctness and parallel execution of the complex bounding overwatch tasks by the MRS. A computational trust model is then constructed by referring to the traversability and line of sight of robots in the terrain. The trust model predicts the trustworthiness of each robot team’s potential behavior in executing a task plan. Robot simulations in ROS Gazebo are used to demonstrate the effectiveness of the proposed framework.
Novel Algorithms for Multi-Agent Autonomous Telerobotic Surveillance and Reconnaissance System
Abstract. The advent of network-based telerobotic cameras enable multiple autonomous agents in the battlefield to interact with a remote physical environment using shared resources. It provides large streams of information (videos or images) to decision makers to conduct military operations such as intelligence, surveillance and reconnaissance, in harsh and hostile environments where it is tedious for humans to collect information. However, the telerobotic cameras system requires a huge amount of data processing and storage units, despite the possibility of redundant (or overlapping) and unimportant information provided by the cameras. To efficiently manage this information using limited resources such that a subset of the information with maximum priority is captured, we introduce a combinatorial optimization problem, referred to as Optimal Target Positioning (OTP), and present exact and approximation algorithms for solving it. More specifically, we prove theoretical properties that significantly reduce the search space for solving this NP-hard problem, and present a customized branch-and-bound exact algorithm to solve it. We also develop a polynomial time approximation algorithm and showcase that the solution value corresponding to its solution is within a factor of 1 - 1/e of the optimal solution value where e is the base of natural logarithm. We present results of our computational experiments conducted to evaluate the performance of our proposed exact and approximation algorithms.
October 16, Friday, 9:30-11:00am eastern time
ARC Cluster of Activities at Clemson University
Dr. Denise Rizzo, S&T Fellow - Office of the Chief Scientist, GVSC.
Dr. Zoran Filipi, Automotive Engineering Chair and VIPR-GS Executive Director,
Dr. Umesh Vaidya, Professor of Mechanical Engineering,
Dr. Robert Prucka, Kulwicki Endowed Professor of Automotive Engineering,
Dr. Beshah Ayalew, Professor of Automotive Engineering,
Dr. Joshua Summers, Professor of Mechanical Engineering,
Dr. Gregory Mocko, Associate Professor of Mechanical Engineering,
Dr. Cameron J. Turner, Associate Professor of Mechanical Engineering, Clemson University
A central theme of the efforts of rapid modernization of the U.S. Army includes virtual prototyping supporting the rapid transformation of U.S. Army fleets. This requires advanced modeling and simulation combined with unprecedented collaboration across disciplines, departments, and facilities. A cluster of ARC projects have started this year at Clemson University to address this issue. The research in this cluster focuses on autonomy-enabled ground vehicles, including digital engineering, energy systems, and human-autonomy teaming in unknown off-road environments. Six exploratory and translational projects have been stared. This presentation will introduce these efforts with the goal of eliciting synergies with other ongoing ARC efforts. Specifically, the following projects will be discussed:
Deep Reinforcement Learning Approach to CPS Vehicle Re-envisioning (ARC Thrust Area 1): Modern-day cyber-physical ground vehicle systems permit superior performance (reconfigurability, robustness, reliability) by exploiting underlying capabilities offered by multi-modal sensor-actuator networks mounted on the electromechanical vehicle bases and orchestrated by algorithmic-intelligence. We explore a Deep Reinforcement Learning enhanced Decision-Support framework to empower superior mobility and information gathering in a range of outdoor terrains.
Integrated Transient Control and Thermal Management of Autonomous Off-Road Vehicle Propulsion Systems (ARC Thrust Area 4): This research focuses on real-time optimization strategies that account for individual component and system response and ensure fast and efficient torque delivery and high-quality electrical power within the thermal constraints of the powertrain. Of particular focus is the management of components to lower thermal footprint while meeting powertrain objectives, thereby minimizing needed package space and cooling requirements. The control methodologies developed will take advantage of forward-looking information, when available from autonomous sensing systems, to better optimize powertrain efficiency, cooling, and electrical energy delivery.
Energy Management of Multi-Scale Vehicle Fleets (ARC Thrust Area 5): This project’s objective is to research and develop energy sharing strategies for mixed fleets of vehicles of varying scales (UGVs and UAVs) operating in a resource-constrained environment. It specifically considers the optimal design and operation of mobile/movable microgrids involving diverse energy sources. It explores modeling and computational schemes for robust and optimal energy utilization plans for UGVs and UAVs.
Computational Representation and Analysis of Mission and System Requirements (ARC Thrust Area 5): The goal of this project is to develop computational reasoning tools to aid in the definition of system-level requirements. The current requirements modeling and management approaches at GVSC and in industry will be studied to identify opportunities for developing new reasoning tools, not currently available. We will first target change prediction and robustness assessment of requirements using established approaches for historical-based modeling and reasoning using networks of requirements. To support this activity, a requirement extraction and definition tool is being developed that can read requirements documents, filter individual requirements, and link requirements into a network of the specifications. Future work will include exploring requirement target setting through systemized gamification of mission simulation, developing a formal logic to requirements, and developing machine learning to identify missing requirements.
Model Interface Specification and Environment to Support Model Integration (ARC Thrust Area 5): The development of next-generation ground vehicles requires the use of models across multiple disciplines, domains, organizations, and software environments. While this is a necessary characteristic of the vehicle development process, it causes information exchange and model integration challenges. To alleviate the challenges associated with model reuse, composition and integration, an ontological approach and the associated model integration framework will be researched, and recommendations will be made to overcome current GVSC challenges. This goal will be achieved by 1) developing an approach to catalog simulations and analysis, 2) formalize a standardized model interface specification, and 3) evaluate existing model integration frameworks.
Best Practices for Computational Tradespace Exploration, Analysis and Decision-Making Tradespace Exploration, Analysis and Decision-Making (ARC Thrust Area 5): In this project, we will refine the overall requirements by establishing project viability in technological, risk, and budgetary spaces. Our approach to enhancing these capabilities include incorporation of tradespace characterization metrics, technical maturity models, Bayesian estimation of requirement prioritization and thresholding, and the demonstration of best-practices for human-in-the-loop decision-making practices for tradespace analysis.
These projects support a collaborative Ground Vehicle Alliance that includes the Clemson Virtual Prototyping of Ground Systems (VIPR-GS) center and the Autonomous Vehicle Mobility Institute (AVMI) at the University of Alabama Birmingham.
November 13, Friday, 9:30-11:00am eastern time
Remote connection details via BlueJeans. Contact William Lim (firstname.lastname@example.org)
Novel Data-Driven Algorithms for Autonomous Vehicle Path Planning Problems with Uncertain Data Parameters
Despite the numerous advantages of ground vehicles (GVs), their size and limited payload capacity lead to fuel constraints and therefore, they are required to make one or more refueling stops in a long mission. Moreover, these operations encounter unknown terrain or obstacles, resulting in uncertainty in the fuel (or time) required to travel among different points of interests (POIs); for example, in a hostile terrain with improvised explosive devices (IEDs), conducting anti-IED sweeps and explosive ordinance disposal can lead to unexpected delays for GVs. In many applications, even the locations of the POIs are not precisely known (uncertain) due to inaccurate a-priori map or imperfect and noisy exteroceptive sensory information or perturbations. This research focuses on the following fundamental question that arises during the planning and execution/operational stages of deploying a team of ground vehicles: How to guarantee a set of paths for heterogeneous GVs with stochasticity in their availability and environment? Specific focus will be on missions with differing unmanned ground vehicle (UGV) and manned ground vehicle (MGV) interactions, and the tasks where the team of UGVs and MGVs need to visit a collection of POIs in the presence of uncertain fuel/time/distance/information required to travel between POIs or reach refueling station.
Resilient Teaming: Fleet organization and decision making in heterogeneous vehicle teams
As autonomy becomes more pervasive, the capabilities of different vehicles may become more pronounced as resources are designed to maximize mission performance, while taking into consideration other requirements such as minimizing energy costs. This introduces an important research problem: How does the configuration of heterogeneous vehicle teams change as autonomy becomes more pervasive in uncertain environments? Methods to define the unique requirements of the fleet across a mixture of vehicles must be developed. Once the make‐up of the team has been determined, a dynamic understanding of the team’s capacity to make real‐time decisions based on the team’s current state, environmental conditions, and uncertainties must be derived. This talk aims to highlight our approach for addressing these needs through resilient teaming: a fleet organization and decision‐making strategy for designing heterogeneous teams to meet these objectives.
December 4, Friday, 9:30-11:00am eastern time
Remote connection details via BlueJeans. Contact William Lim (email@example.com)
Dynamic Task Allocation and Understanding of Situation Awareness Under Different Levels of Autonomy in Closed-Hatch Military Vehicles
PI: Dr. Cindy Bethel, Associate Professor of Computer Science and Engineering, and the Billie J. Ball Endowed Professor in Engineering, Mississippi State University
Co-PI: Dr. Daniel Carruth, Graduate Ph.D. Candidate: Jessie E. Cossitt
Link to project
With current autonomous vehicle capabilities, it is necessary for operators to remain engaged and monitor the system, intervening when necessary. This creates a need to better understand the interactions between operators and autonomous vehicle control systems in order to provide the best-case scenario for utilization of autonomous capabilities. Such an understanding could lead to the development of a system to dynamically allocate tasks in military missions to reduce crew sizes and thus reduce labor costs. The goal of this research is to determine how increasing levels of autonomous capabilities in vehicles affect the operator’s situational awareness, cognitive load, and performance responding to road events as well as responding to other auditory and visual tasks. Understanding the interactions among these factors is necessary to eventually determine the best way to allocate tasks to crew members in missions where crew size has been reduced due to the utilization of autonomous vehicles.
Optimal Distribution of Tasks in Human-Autonomy Teams
Bogdan Epureanu (PI, Professor), Amin Ghadami (Post-Doc), Haochen Wu (Grad. Student), UM
Emrah Bayrak, Stevens Institute of Technology
Victor Paul (Team Leader, Motion Base Tech.), John Brabbs (Branch Chief - Software Development, Software Engineering Center), Jillyn Alban (Research Engineer), GVSC
Mert Egilmez (Mechanical Engineer), Veoneer-Nissin (now ZF)
Link to project
Humans and autonomous vehicles can collaborate as a team to improve capabilities of existing human-operated systems. Autonomous vehicles are capable of handling dangerous tasks but limited in real-time computational power, perception accuracy, and making ethical decisions, while humans are more adaptive and creative problem-solving skills but are limited in terms of cognitive loads and fatigue. Careful integration of humans and autonomous vehicles enables teams to survive and function in complex environments by combining the efficiency of autonomy and the characteristics of humans. This project presents a framework for autonomous vehicles to learn collaboration with humans considering human factors in the learning process. A decentralized and adaptive planning algorithm under dynamic demands and uncertainties is proposed to give equal authorities to agents during operations. Using the developed algorithm, we investigate how to distribute optimally tasks among autonomous vehicles and humans in complex operations that require the execution of multiple tasks simultaneously, and how to train autonomous vehicles to collaborate with humans to complete missions. In addition, we elucidate the benefits and limitations of humans in the team with example scenarios and investigate the effect of heterogeneity in the capabilities of human and autonomous vehicles on the team performance.