Skip to content
vipr-gs-prototype
VIPR-GS

Virtual Prototyping of Autonomy-Enabled Ground Systems (VIPR-GS)

Research Focus Area:

Off-Road Autonomy

The autonomy area focuses on the fusion of data from sensing and estimation to provide situational intelligence and machine learning for algorithm development and novel approaches to multi-scale team routing.

Autonomous ground-combat systems capitalize on enhanced scientific understanding and ongoing technological revolution to augment multi-domain operations' mobility and maneuverability components. The goal is to engineer configurability, resilience, and graceful degradation in the context of systems of automated vehicles.

Focus Area Director

Yunyi Jia
McQueen Quattlebaum Associate Professor, Department of Automotive Engineering
yunyij@clemson.edu

vipr-research.jpg

Overview

Developed capabilities enhance all steps of the Observe, Orient, Decide, and Act framework:
  1. Observe
    • Sensor pose control: We developed pose control and optimization of sensors onboard autonomous vehicles to provide the best orientations to take measures of the environment. 
    • Sensor data characterization: The team performed noise filtration and processed LiDAR data in real time under foggy and no-fog conditions.
  2. Orient
    • Sensor simulation & navigation: AI/ML support. Simulation to create ML models that work when sensors are degraded
  3. Decide 
    • Multi-modal human-machine interface: We have developed a theoretical framework that captures how human-agent teammates should mutually collaborate within tasks. Based on the theoretical framework, we developed and evaluated prototypes of a human-centered multimodal system interface in a virtual reality environment
    • Shared control: To combine human control with autonomy, we developed shared control strategies integrating deep reinforcement learning and imitation learning techniques.
  4. Act 
    • Physics-based modeling: created unique models for tracked vehicles (currently transitioning to GVSC – Matthias/Qulin)
    • Physics-based control: individual vehicle control based on vehicle physics

Research Efforts

  • 1.1 Deep Reinforcement Learning Approach to CPS Vehicle Re-envisioning

    Principal Investigator

    Venkat Krovi (vkrovi@clemson.edu)

    Motivation

    Re-envision autonomy with Cyber-Physical System (CPS) off-road vehicle architecture concepts

    Goal

    Extract robust and reliable mobility & maneuver performance

    Approach

    AI-based frameworks to support both offline design-selection and online dynamic-orchestration of hierarchies of parameters, settings and behaviors

  • 1.2 Enhanced Situational Intelligence for Off-Road Depot Vehicle through Collaborative Perception and Human-centered Algorithmic Intent

    Principal Investigator

    Sophie Wang (yue6@g.clemson.edu)

    Motivation

    Performing situational intelligence in terms of correlating, analyzing and visualizing disparate data, and providing intelligent decisions and actions out of a massive amount of data, is one of the major critical functions of armaments and military equipment

    Goal

    Provide situational awareness of both the mission environment and the human operator for effective automatic/manual decision-making inside a mobile depot vehicle

    Approach

    • Environmental situational intelligence: Sensor uncertainty characterization, sensor pose optimization, sensor fusion, semantic mapping, surrounding entities modeling and prediction
    • Human situational intelligence: (Inverse) reinforcement learning, automated decision-aid, prototype of human-computer interface
  • 1.3 Sizing, Routing, and Coordinating Multi-Scale Ground Vehicle Fleets

    Principal Investigator

    Pamela Murray-Tuite

    Motivation

    Develop techniques for smart management of forward-deployed fleets for mobility in unstructured environments

    Goal

    To determine the sizing, routing, and coordination of a heterogeneous team of AGVs in unstructured environments with AGV losses

    Approach

    Mathematical approaches and algorithms for adversarial reasoning, threat detection, team sizing, routing, distributed coordination, physics-based estimation and control

  • 1.4 Cross-Cutting Autonomy-Enablers

    Principal Investigator

    Richard Brooks

    Motivation

    • Vehicle fleets are complex systems of systems​
    • Fleet needs: learning, security, connectivity, error detection, integration​
    • Framework: Correct failures, tolerate attacks, create resiliency

    Goal

    • End-to-end mission coordination​
    • System aspects needed for distinctive GVSC mission​
    • Analyze performance, security, and adaptation

    Approach

    • Instrumented testbed in simulation and with vehicle platforms ​
      • Accessible by VIPR GS researchers ​
      • Evaluation of an autonomy architecture and the supporting functions.​
      • Agile, adaptive, responsive task authorization using Zero Trust overlay for ROS-like middleware. ​
    • Adaptive UGV collision prediction and avoidance using Markov Chain with CARLA simulator and fault detection using Hidden Markov Models ​
    • ROS 1 and ROS 2 network tests for drones and simulated vehicle fleets​
    • Developing the software-defined radio platform for dynamic spectrum access and customized communication interfaces.​
  • 1.5 Efficiency-Based Fleet Traversing Scheme Design Under the Risk of Disruptions Using Goal Programming Model

    Principal Investigator

    Judith Mwakalonge

    Motivation

    Supply chain disruption has become global due to the pandemic and the Russia-Ukraine war, which can result in severe economic and financial consequences. Route disruptions imply significant breakdowns in distribution nodes that comprise a supply chain. The design of an efficient and resilient fleet traversing scheme would be essential for the strategic plan of logistics systems .

    Goal

    Demonstrate how to design efficient fleet routing/traversing schemes under the risk of route disruption, given a set of transportation requests, a fleet of vehicles with capacities, and origin and destination nodes while simultaneously considering the cost/time-efficiency of the system

    Approach

    Considering three objectives simultaneously, a goal programming (GP) approach with multi-objective functions is applied to formulate an optimization model. For evaluating various vehicle traversing schemes generated by GP, the data envelopment analysis (DEA) methods are applied to identify efficient traversing schemes.
  • 1.22.6 Open-Autonomy Verification & Validation (V&V) Framework

    Principal Investigator

    Venkat Krovi

    Motivation

    Innovative Verification & Validation (V&V) approaches to evaluate autonomy algorithms for offroad CPS ground vehicles​

    Goal

    Refinement of a candidate ADAS algorithm (semantic segmentation to support motion re-planning) with variability included​

    Approach

    Innovations build upon coupling of:​
    • Novel scientific methodologies (sequential sampling; AI surrogates-models, Digital Twins)​
    • AI-based testing workflows (semantic Scenario Description Languages to guide real-time V&V for extended “software algorithm-CPS asset” model) ​
    • Enterprise-scale tools and methodologies (Simulation-as-a-Service; containerized cloud-based workloads) to manage the big-data lifecycle from cradle to grave​
  • 1.22.7 Investigation of emulators for Single Photon Lidar to determine its suitability for autonomous vehicle integration

    Principal Investigator

    Goutam Koley

    Motivation

    Imaging hidden objects behind semipermeable barriers like tall grass and light foliage is very important for vehicle autonomy and safety

    Goal

    To investigate the capability of a Linear Mode Lidar (LML) on a ground vehicle to perform imaging of hidden objects behind semipermeable barriers emulating the performance of a Single Photon Lidar (SPL)

    Approach

    Use an LML mounted on a ground vehicle to capture multimodal data, develop image processing software to extract hidden objects behind semipermeable barriers, and finally determine the feasibility of using SPL in a ground vehicle for hidden object imaging​
  • 1.22.8 Determining Soldiers’ Workload while Operating Ground Vehicles

    Principal Investigator

    Johnell Brooks

    Motivation

    Evaluate the relationships between physiology, perceived workload levels, and operator performance in Army drivers.

    Goal

    Develop an algorithm that may predict operator states in real-time.

    Approach

    Clemson University, the Ground Vehicle Systems Center (GVSC), the U.S. Army Aeromedical Research Laboratory (USAARL), the Army Research Lab (ARL), the Edward Hines Jr. VA Hospital, and the South Carolina Army National Guard (SCARNG) are collaborating to address this important topic.
  • 1.22.9 Spatial-AI Real-Time Mapping for Off-Road Ground Vehicles

    Principal Investigator

    Bing Li

    Motivation

    • Spatial AI mapping enables the off-road vehicle to accurately perceive and understand complex and challenging environments.​
    • Spatial AI in real-time design provides new opportunities to learn critical in-situ environment and vehicle conditions for efficient navigation.

    Goal

    • Real-time mapping algorithms that utilize multi-modality sensors produces map representation for navigation.​
    • Spatial AI models for sensing, terrain moisture predictions, and mapping construction.

    Approach

    Real-time simultaneous localization and mapping (SLAM) with map abstractions augmented by backend spatial AI learning and recognition.
  • 1.23.10 Monitoring and maintaining trustworthy networked autonomy in a zero trust environment

    Principal Investigator

    Fatemeh Afghah

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come
  • 1.23.11 VANTAGE: Vehicular Aerial Navigation of Tethered Autonomous Ground Systems

    Principal Investigator

    Matthias Schmid

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come
  • 1.23.12 Standardized modular secure firmware update framework for military vehicles

    Principal Investigator

    Mert Pese

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come
  • 1.23.13 PRECOgniTION: PRobabilistic prEdiction from CONtext determinaTION

    Principal Investigator

    Matthias Schmid

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come
  • 1.23.14 Off-Road Obstacle Detection Analysis for Autonomy-Enabled Ground Vehicle Navigation

    Principal Investigator

    Judith Mwakalonge (SCSU) and Yunyi Jia (CU)

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come
  • 1.23.15 Virtual Sensor Reconstruction for offroad autonomous vehicle

    Principal Investigator

    Feng Luo

    Motivation

    More info to come

    Goal

    More info to come

    Approach

    More info to come