12/11/2020 Dr. Judy Hoffman discussed the spectrum of dataset and model bias ranging from inadvertent visual changes to adversarially manipulated images. She then covered techniques for bias mitigation, including domain adversarial learning, which facilitates transfer of information between different visual environments and across different semantic tasks thereby enabling recognition models to generalize to previously unseen worlds, such as from simulated to real-world driving imagery. Finally, she touched on the pervasiveness of dataset bias and how this bias can adversely affect underrepresented subpopulations
12/04/2020 Dr. Nicolas Norboge provided an overview of transportation funding modeling in the era of COVID for the nearterm, but also long-term impacts such as those associated with connected and automated vehicles.
11/20/2020 Dr. Cynthia Rudin presented several stories that provide interesting lessons for modern design of models, the ways we look into them, and sacrifices we might need to make when we try to look at high dimensional data
11/13/2020 Dr. Bates gave a brief overview of a novel active-learning technique for fault diagnosis of a partially-known system modeled as a finite-state Discrete Event System (DES). The proposed technique first tabularly gathers information from a diagnosis tool, termed diagnoser, constructed from the available information of the known part of the system. An active-learning technique is then executed to iteratively capture information regarding the unknown portion of the system to construct the remaining part of the diagnoser. The final constructed diagnoser is ultimately able to detect and identify occurred faults through the examination of the observable behavior of the system. In order to display the details of this method, we have applied this method to a real-world case study concerning the Boeing 737-MAX and the implementation of the Maneuvering Characteristics Augmentation System (MCAS).
11/06/2020 Mr. Navarro gave a brief overview of a time-critical cooperative framework for autonomous mobility of a fleet of heterogeneous vehicles operating in challenging scenarios. Multi-agent motion planning in cluttered scenarios is a challenging and resource intensive problem. To mitigate this, he reexamined the backbone of path-planning algorithms, the proximity queries that determine whether a path is collision free; leveraged silhouette information from nearby obstacles to expedite solutions through narrow passages; and smoothened the resulting solution to meet desired continuity and differentiability requirements. He introduced a non-linear path-following algorithm with guaranteed performance bounds for the execution of the smooth 4D trajectories.
10/30/2020 Dr. Chase discussed the concepts and standards under development in Connected Traffic Signals, descriptions of pilot deployments around the US, and introduced sample data for researchers interested in exploring this field. Connected Traffic Signals are just one component of Connected and Autonomous Vehicle systems; however, they are at a critical juncture between manufacturers of Autonomous Vehicles and infrastructure maintaining agencies like Departments of Transportation.
10/16/2020 Dr. Ezer presented the challenges and opportunities for HMT across civilian and military operations. Examples included identifying anomalous vessel behaviors in a maritime environment, supevisory control of swarms of unmanned vehcles and adaptive human-AI interactions based on wearable sensors.
10/09/2020 Dr. Brian German presented recent research focused on the conceptual design, analysis and operations of electric takeoff and landing (eVTOL) aircraft for urban air mobility (UAM). Specific topics include the development of a battery model appropriate for aircraft sizing and an investigation of the flight performance of canonical eVTOL aircraft configurations.