ARC Collaborative Research Seminar Series
Fall 2017

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

Remote attendance via tele/video conference: Contact William Lim (

Refreshments will be served 9:15-9:30am. The talks will begin at 9:30 a.m. sharp.

Event venue alternates between University of Michigan (Ann Arbor) and U.S. Army TARDEC (Warren)

September 8, Friday (9:30a.m. - 11a.m.)
University of Michigan, UM North Campus, Duderstadt Center. room 1180

1. Algorithms for Coordination and Collaboration in Heterogeneous Unmanned Systems
Dr. Pratap Tokekar, Asst. Prof. Electrical and Computing Engineering, Virginia Tech

Abstract & Biography         The technology enabling unmanned systems is becoming mature enough to enable operations in the field. The next frontier is enabling a team of unmanned vehicles with limited human oversight. What is needed are algorithms that can enable effective coordination and collaboration within a team of heterogeneous robots and possibly humans. In this talk, I will present our recent work on devising efficient algorithms with provable performance guarantees for multi-robot coordination and collaboration.
        We will focus on the routing and coordination of aerial and ground robots. I will present approximation algorithms for multi-robot teams to map known and unknown environments in the least amount of time. I will then present a heterogeneous formulation for the classical Traveling Salesman Problem and show how to apply it to plan for aerial and ground robots. Finally, I will describe our new work on informative path planning for coordinated sampling in marine environments with aerial robots and robotic boats.
        Pratap Tokekar is an Assistant Professor in the Department of Electrical and Computer Engineering at Virginia Tech. Previously, he was a Postdoctoral Researcher at the GRASP lab of University of Pennsylvania. He obtained his Ph.D. in Computer Science from the University of Minnesota in 2014 and Bachelor of Technology degree in Electronics and Telecommunication from College of Engineering Pune, India in 2008. He is a recipient of the NSF CISE Research Initiation Initiative award. His research interests include algorithmic and field robotics, and cyber physical systems, and their applications to precision agriculture and environmental monitoring.

2. Formal Methods for Provably-correct Control Design for Cyber-physical Systems
Dr. Necmiye Ozay, Asst. Prof. Electrical Engineering and Computer Science, University of Michigan

Abstract         Formal methods are mathematical techniques, originally proposed by the computer science community, to rigorously analyze software systems. In recent years we have witnessed an increase in the use of techniques originating in this area to solve control problems for cyber-physical systems. Similarly, the idea of automatically synthesizing a controller that enforces the desired specifications is becoming an alternative to the verification paradigm prevalent in the formal methods area. This talk will provide a brief introduction to formal verification and synthesis for control systems. By using some autonomous cyber-physical system case studies my group works on, I will illustrate how these techniques can be used to design provably-correct controllers, find corner cases in existing designs and characterize safe operating envelopes and robustness-performance trade-offs.

September 22, Friday (9:30a.m. - 11a.m.)
University of Michigan, UM North Campus, GG Brown Bldg, room 2540 (Grand Conference Room)

1. Fast Numerical Algorithms for High-fidelity Simulation of Terramechanics (ARC project)
Dr. Shravan Veerapaneni, Assoc. Prof. Mathematics, University of Michigan

Abstract The discrete element method (DEM) is one of the most widely used approaches to simulate the dynamics of multiple rigid bodies, specially in terramechanics applications. While the classical DEMs are based on applying penalty forces to handle contact, novel complementarity constraint based approaches (DEM-C) are emerging as promising alternatives that ameliorate the numerical stability issues plaguing the penalty methods. Motivated by prior work at TARDEC, we are interested in developing fast and scalable algorithms for solving the constrained optimization problem that arises in DEM-C. I will discuss the progress by our group in the past year, the challenges that lay ahead and the broader impacts of this direction of research.

2.Towards Kinematic Reconstruction of Roadway Scenes from Single-Camera Input
Dr. Jason Corso, Assoc. Prof. Electrical Engineering and Computer Science, University of Michigan

Abstract Creating simulation environments takes ample well-annotated data. The recent trend moving toward machine learning-based construction and use of simulations has increased the need for such data. However, acquiring such well-annotated data is a challenge, especially in the case of rare traffic events like crashes. Surprisingly, web-uploaded open-source videos acquired from simple devices like dash cameras already contain ample examples. Yet, their 3d kinematic and dynamic parameters are not known. In this talk, I will describe our recent efforts in inferring such parameters for third-person view dash-camera acquired video. Our works implement human-in-the-loop deep network architectures to reconstruct vehicle pose from a single viewpoint and to request input from a human to help with this process.

October 20, Friday (9:30a.m. - 11a.m.)
U.S. Army TARDEC, 6501 E. 11 Mile Road, Warren, MI 48397-5000
Building 200B TARDEC University Class Rooms A&B

ARC projects presenting:
1. Situational Awareness, Trust in Autonomous Vehicles and Secondary Task Performance
Dr. Lionel Robert, Asst. Prof. School of Information, University of Michigan

Abstract Autonomous vehicles (AV) provide drivers with the ability to fully immerse themselves in a secondary task. Unfortunately, drivers may not trust the vehicle, which prevents them from either handing over the driving or fully focusing on the secondary task when they do. In the first case the driver fails to complete the secondary task while in the later they face cognitive overload which decreases performance in the secondary task. We assert that situational awareness (SA) or a lack of it is the primary problem. Trust in the AV is a function of its capability relative to the driving demands in a given situation. However, driving is often dynamic and unpredictable. This makes it difficult for the driver to focus on the task and ensure that the driving situation has not exceeded the vehicle’s capability. Increasing the driver SA should increase their trust in the AV, decrease their mental workload and lead to better secondary task performance. We conducted a human-in-the-loop study and manipulated SA and assigned the participant a secondary task. Data collected included measures of trust, trust behavior, and mental workload in the form of survey, eye-tracking, and heart rate data. Preliminary results show that SA promoted AV trust, reduced mental workload and led to better secondary task performance.

2. Novel Hybrid Electric Powertrains Enabled by Models of Electro-Magnetic-Structural Dynamics
Dr. Bogdan Epureanu, Prof. Mechanical Engineering, University of Michigan
Dr. Heath Hofmann, Assoc. Prof. Electrical Engineering, University of Michigan
Chenyu Yi, Mohammad Khodabakhsh, Graduate Students

Abstract The increasing demand for lower emissions and higher fuel efficiency requires the design and optimization of hybrid electric architectures which reduce noise while reinforcing the performance of powertrains and electro-magnetic-structural systems. Design and optimization of powertrain architectures requires accurate yet computationally efficient models to fulfill multiple tasks such as vehicle driving simulations and structural analysis. In addition to the vibratory electromagnetic forces created on the structure of the electric machines, other periodic excitations are significant also in most hybrid electric powertrains due to torque pulsations in designs that have internal combustion engines with few cylinders. These significant torque fluctuations can cause deformation in the structure of the electric machine. As a consequence, mechanical and electromagnetic features are coupled. Previous studies do not focus on capturing the coupling between electric machine forcing and deformation of the structure with cyclic asymmetry (mistuning). In this project, a novel hybrid electric powertrain architecture was designed, modeled, and optimized. The construction of a structural reduced-order model considering small mistuning, the computation of model electromagnetic force, and the methodology of integrating the electromagnetic force with the structural reduced-order model is presented.

November 17, Friday (9:30a.m. - 11a.m.)
U.S. Army TARDEC, 6501 E. 11 Mile Road, Warren, MI 48397-5000
Building 200B TARDEC University Class Rooms A&B

ARC projects presenting:
1. Intelligent Reconnaissance: An Energy‐Efficient Method for Multi‐Robot Reconnaissance in an Unknown Environment (project link)
Dr. Kira Barton, Asst. Prof. Mechanical Engineering, University of Michigan

Abstract In off-road environments, robots may traverse a variety of terrains for which the energy cost of traversal is unknown. This work focuses on the development of a probabilistic method for predicting energy usage of ground robots across spatially varying terrains. To address this problem, we aim to build a spatial energy map of the terrain, modeled with Gaussian process regression (GPR), based on energy usage from data collected as a robot traverses the environment. The probability distribution of a new path’s energy cost is computed by integrating over the model. To incorporate energy predictions into decision-making, a chance-constraint is provided for a discretized graph-based environment. A heuristic algorithm is developed to ensure that a robot only moves to locations in which it is feasible to return to a recharging base without depleting its energy. To demonstrate the application of this work, we apply energy mapping to a multi-robot system to identify energy-constrained reachable locations in the environment, based on the locations of recharging bases. Bounds are proven in a deterministic case to describe the reachability of the multi-agent system as compared to the true reachable set. Experimental validation of the path energy prediction method is conducted on a robotic platform.

2. Situational Awareness and Sustained Survivability through Man/Unmanned Teaming (project link)
Dr. Dimitra Panagou, Asst. Prof. Aerospace Engineering, University of Michigan

Abstract The SQUAD project is motivated by the need for protection of Ground Vehicle systems against Aerial Threats, and in particular, against small Unmanned Aerial Vehicles (UAVs). It is thus of vital importance that GVs are equipped with systems that can detect and act against small UAVs. In this talk we will review our algorithmic developments and experiments towards Aerial Sensing/Surveillance around Ground Vehicles, and our ongoing work on resilient swarms of Autonomous (Ground and Aerial) Vehicles, towards adversarially robust autonomous systems.

ARC members can download the presentation files on our password-access online portal iARC.
Non-ARC members please email with your requests.