Skip to main content
arc logo
Automotive Research Center
hero image

2019 ARC Research Seminars - Fall Series

ARC seminars are free and open to the general public. Center members can download the presentation files on our password-access online portal iARC. Non-ARC attendees please email with your requests.

Parking & directions inquires: Contact by 2:00 p.m. the day before the seminar, or look up campus visitor parking locations.

Remote attendance via tele/video conference: Contact William Lim

September 27, Friday, 9:30-11:00 AM eastern time

University of Michigan, UM North Campus, Phoenix Memorial Lab. room 2000A

Communication-Constrained Multi-Robot Coordination

PI: Prof. Edwin Olson, Computer Science and Engineering, U of M

Abstract: We propose an approach to allow a team of collaborative robots to dynamically select a policy (a “strategy”) that takes into account the state of robot team, the amount of available bandwidth, and the configuration of the current environment. In particular, we describe how coordinating robots can play “tag” with an evader, dynamically changing their strategy based on those factors. The approach is based on Multi-Policy Decision Making, in which the robots run Monte Carlo simulations based on their current estimate of the world, trying a different coordination strategy on each Monte Carlo run. The approach allows the robot to dynamically adjust its behavior without the system designers having to anticipate and pre-program the best strategy for every situation.

Link to Project

Adversarially Robust Coordination for Autonomous Multi-Vehicle Systems

PI: Prof. Dimitra Panagou, Aerospace Engineering, U of M

Abstract: This ARC project aims to increase autonomy of networked multi-vehicle systems in adversarial environments by considering both the safety and the resilience aspects of dynamic multi-agent networks. Safety refers to the generation of guaranteed collision-free trajectories for multiple vehicles (agents) so that they navigate efficiently in cluttered environments while collaborating towards a common task (e.g., data gathering). Resilience refers to the guaranteed safe accomplishment of the mission despite the presence of possible adversaries that can send malicious data over compromised communication links. In this talk we will present our results on resilient communication structures, and on adversarially robust controllers that maintain safety of the individual agents, filter out the effects of adversarial agents, and guarantee mission accomplishment.

Link to Project

October 25, Friday, 9:30-11:00 AM eastern time

University of Michigan, North Campus, GG Brown Lab. room 2540 (Grand Conference Room)

Physics‐Based Multiscale Continuum‐Discrete Deformable Terrain Model for Off‐Road Mobility Simulation

PI: Prof. Hiroyuki Sugiyama, Mechanical Engineering, U. of Iowa

Abstract: This talk presents a high-fidelity hierarchical multiscale off-road mobility simulation capability to address limitations of existing computational deformable terrain models, including the use of phenomenological constitutive assumptions in finite element (FE) approaches and high computational intensity of discrete element (DE) models. We will discuss a scalable parallel computing scheme for the multiscale off-road mobility model, composed of a detailed multibody vehicle, nonlinear FE tires, and the upper-scale FE terrain integrated with the lower-scale DE representative volume elements through the scale bridging algorithm. We will also present the comparative performance evaluation as well as the full-scale vehicle validation against test data of the Next Generation NATO Reference Mobility Model (NG-NRMM) Cooperative Demonstration of Technology (CDT).

Link to project

Fast Numerical Algorithms for High-fidelity Simulation of Terramechanics

PI: Prof. Shravan Veerapaneni, Mathematics, U of M

Abstract: In this talk, we will present our recent work on efficient, hybrid MPI/OpenMP framework for the cone complementarity formulation of large-scale rigid body dynamics problems with frictional contact. Parallelization of the state-of-the-art first and second-order optimization solvers will be discussed. Our approach is highly scalable: a sedimentation simulation involving 256 million particles (~324 million contacts on average) was resolved using 512 cores in less than half-hour per time-step on the FLUX system. Then, we will discuss our ongoing efforts on compressing the generated simulation data using tensor decompositions to help in rapid construction of mobility maps using machine learning approaches.

Link to project

November 22, Friday, 9:30-11:00am eastern time

University of Michigan, North Campus, GG Brown Lab. room 2540 (Grand Conference Room)

Intuitive Interfaces to Support Control Authority Exchanges between Driver and Automation

PI: Prof. Brent Gillespie, Mechanical Engineering, U of M

Abstract: A human driver plays critical roles even in conditionally automated and highly automated driving systems. The driver directs and supervises the automation, but also covers for situations that exceed automation design limits. To ensure smooth transitions when the driver must take over control with little or no warning, the interface to the automation system (and to the vehicle) must be intuitive. Haptic shared control is designed to support fast and intuitive transitions of control authority by conferring authority to the automation system in a manner that mimics cooperative manipulation between two humans. Both the driver and automation express and read the other’s control actions (steering angle command) and control authority (impedance) through a steering wheel that is directly coupled to the tires. However, two cooperating humans will typically supplement their pushing and pulling with communication in other modalities. To set up an additional two-way communication pathway, we are prototyping a grip-force sensing and shape-changing steering wheel. Since grip force typically varies directly with the driver’s impedance, an automation system that senses grip force can track the driver’s desire to take over control and can yield when appropriate. If the situation instead warrants an automation takeover, the automation can inform the driver by changing steering wheel shape. We have developed a system model and prototype for a pneumatically-driven and pressure-sensing steering wheel that describes two-way haptic communication in the axis of grip. We are using this steering wheel to supplement two-way communication in the axis of control through haptic shared control. We will also describe our experiments designed to validate the model and evaluate its utility in a simulated driving environment.

Link to project

Modeling Bi-directional Trust in Semi-autonomy for Improved System Performance

PIs: Prof. Dawn Tilbury, Mechanical Engineering and Prof. Lionel Robert, School of Information, U of M

Abstract: Trust in automated driving systems (ADSs) is a key factor for understanding whether drivers choose to leverage their capabilities. With this project, we aim to investigate factors that impact drivers’ trust in ADSs and model how trust develops over repeated driver-ADS interactions, with the ultimate goal of enhancing system usage effectiveness. Our approach is based on a series of human-in-the-loop experiments, where participants interact with an ADS in a driving simulator with varying situational awareness, automation reliability, and driving conditions. At the same time, the participants are asked to report their trust in the ADS recursively, while having some of their trusting behaviors monitored. We will present findings on how increased situational awareness, higher automation reliability and the driving conditions influence drivers’ trust in the ADS, and how these findings can be used for the design of adaptive ADSs.

Link to project

December 6, Friday, 9:30-11:00am eastern time

University of Michigan, North Campus, GG Brown Lab. room 2540 (Grand Conference Room)

A Decision-Based Mobility Model for Semi and Fully Autonomous Vehicles

PI: Prof. Vijitashwa Pandey, Industrial and Systems Engineering, Oakland U.

Abstract: With the emergence of intelligent ground vehicles, an objective evaluation of vehicle mobility has become an even more challenging task. Whereas a vehicle’s off-road mobility may be significantly enhanced with intelligence, it also introduces many new variables into the decision making process that must be considered. We present a decision analytic framework to accomplish this task. In our approach, a vehicle’s mobility is modeled using a utility function that captures an operator’s preferences over multiple mobility attributes of concern. This function captures the operator’s tradeoff behavior over the attributes, as well as their risk attitude. We also provide a method to analyze various operating scenarios including the ability to mitigate uncertainty in the vehicles inputs. An example of this is the collection of soil properties data using techniques such as remote sensing. Operators of these vehicles are interested in finding the value of collecting such information. In this talk, we will present some results from our study and discuss the anticipated next steps.

Link to Project

Probability of Mobility for Mission Planning of Autonomous Ground Vehicles at “High Stress” Environments

PIs: Prof. Zissimos Mourelatos, Mechanical Engineering, Oakland U.; Prof. Zhen Hu, Industrial and Manufacturing Systems Engineering, U of M Dearborn

Abstract: Physics-informed modeling and simulation (M&S) is essential for vehicle mobility prediction under various terrain and soil conditions. The presence of natural variability of terrain and soil properties as well as the vehicle system itself, cause variability in vehicle mobility. Due to model simplifications, assumptions, and numerical discretization errors, epistemic uncertainty (lack of knowledge) in M&S adds another layer of uncertainty to predicting mobility in addition to the natural variability. This could lead to large errors in the probability of mobility prediction and thus wrong decision making. While mobility testing data can be used to reduce the mobility prediction uncertainty, it may be difficult and/or expensive to collect. This project aims 1) to improve the prediction confidence of a vehicle mobility prediction model using the least number of mobility tests, and 2) to account for the mobility reliability in mission planning of autonomous vehicles. In this talk, we will present our work on mobility testing design optimization for uncertainty reduction in generating an off-road mobility map using a Bayesian approach. With a focus on reducing the epistemic uncertainty in M&S, our approach is able to considerably increase the prediction confidence of vehicle mobility using a minimum number of mobility tests at “high-stress” environments. We will also discuss our ongoing efforts on model bias correction of a vehicle mobility model by integrating model verification and validation and model calibration.

Link to Project