Title: Towards Decoupling the Safety vs. Performance Tradeoff in Autonomous Driving

Abstract: Ensuring autonomous vehicle safety is challenging due to the need to numerically represent complex dynamics and workspace constraints (e.g., collision avoidance). The challenge is further compounded by the fact that all models are wrong; in other words, we must account for uncertainty in our vehicle's motion and the world around it. This talk focuses on Reachability-based Trajectory Design (RTD), a recent approach to incorporating uncertainty in motion planning. First, we apply RTD on an autonomous vehicle. Then, we show how to exploit reinforcement learning to maximize performance without sacrificing safety.

Bio: Shreyas Kousik is an assistant professor in the George W. Woodruff School of Mechanical Engineering. Previously, Shreyas was a postdoctoral scholar at Stanford University, working in the ASL under Prof. Marco. Kousik completed a postdoc with Prof. Grace Gao in the NAV Lab. He received his Ph.D. in Mechanical Engineering at the University of Michigan, advised by Prof. Ram Vasudevan in the ROAHM Lab and received his undergraduate degree in Mechanical Engineering at Georgia Tech, advised by Prof. Antonia Antoniou.

Title: Semantics-Aware Proactive Informative Planning in Extreme Environments

Abstract: Autonomous systems operating in extreme environments demand high-fidelity decision making capabilities that are resilient to errors from imperfect sensing and can actively respond to diverse, unexpected situations. However, conventional decision making approaches have faced challenges, including degraded perception, risks associated with imperfect controls and forward planning, errors from non-real-time updated models, and brittleness to environmental changes. This talk presents a semantics-aware decision making framework that incorporates proactive informative planning and considers sensing, inference, and acting in planning steps from semantic representation of environments. By leveraging semantics, autonomous systems can achieve human-level understanding and reasoning in decision making, which allows for increased robustness. This talk then introduces a JPL NeBula (Networked Belief-aware Perceptual Autonomy), a robot-onboard autonomy system that utilizes model-based design and data-driven generalization for perceptual representation and decision making. A series of field test results with NeBula for active search for signal-emitting objects in a fully unknown building environment is delivered. Finally, this talk explores potentially applicable domains for this approach in real-world autonomous missions.

Bio:  Sangwoo Moon working with Dr. Ali Agha at NASA Jet Propulsion Laboratory (JPL). Before joining the JPL, he was a research engineer at Texas A&M University working with Prof. Reza Langari and Prof. John Valasek. He received a Doctoral Degree under Prof. Eric Frew at the Ann and H.J. Smead Aerospace Engineering Sciences of the University of Colorado Boulder, May 2021. He was a faculty officer at the Korea Air Force Academy who taught cadets for three years (2011-2014) after he graduated Master’s Degree and worked as a researcher at the Aerospace Research Institute of the Korea Air Force Academy (2014-2015). He obtained a Master’s Degree and a Bachelor’s Degree in Aerospace Engineering from KAIST, South Korea, in 2011 and 2009, respectively.

Title: Context-Aware Planning and Control for Autonomous Urban Driving 

Abstract: Autonomous urban driving has a longstanding promise of deploying fully autonomous taxis and self-driving cars. However, the first step towards full autonomy is integration of the self-driving cars in our existing infrastructure and being able to drive side-by-side human driven cars.  In this talk, I will highlight our work on context-aware path and motion planning for autonomous urban driving. Specifically, how we use the road contextual information, driver intent and driving styles for high-level path planning and low-level controls. I will also present our development, simulation, and deployment efforts of Autonomous Mobility-on-Demand (AMoD) solutions in mixed traffic environments in Singapore.

Bio:  Dr. Malika Meghjani is an Assistant Professor in the Computer Science and Design Pillar at Singapore University of Technology and Design (SUTD). She directs the Multi-Agent Robot Vision and Learning (MARVL) Lab, with the focus on algorithm design for efficient, reliable and scalable robots that can work independently and collaboratively with humans. Her research interests are in planning under uncertainty, reinforcement learning, computer vision, deep learning, and game theory. The applications of her work are in field robotics ranging from marine robots specifically, underwater and surface vehicles to self-driving cars and other ground vehicles in unstructured environments. 

Malika has been cited by Analytics Insight in 2020 as one of the World's 50 Most Renowned Women in Robotics. She is also 2017 SMART Postdoctoral Scholar, 2015 McGill Scarlet Key recipient, 2013 IEEE Canada Women in Engineering Prize awardee and 2013 Google Anita Borg Scholar. 

Title: Leveraging limited sensor measurements beyond the visible spectrum for robust navigation

Abstract: This talk presents an innovative method for robust navigation using limited sensor measurements beyond the visible spectrum, with implications for navigation under challenging and constrained conditions. We propose a novel algorithm that merges data from various non-visible spectrum sensors, including infrared and radio waves. The methodology is underpinned by a fusion approach, which employs advanced machine learning techniques, to better interpret and integrate the data from the diverse sensor readings. Real-world testing scenarios demonstrate the robustness of our proposed system, even in environments with limited visibility such as inclement weather, darkness, or dense smoke. This approach offers significant improvements for applications like autonomous vehicles, mobile robotics, and unmanned systems, where navigation accuracy and reliability are paramount. Our findings advocate for a shift towards an inclusive sensory-data approach for navigation, which could transform current practices and offer new possibilities for maneuvering in challenging environments.

Bio: Ayoung Kim is currently working as an associate professor in the department mechanical engineering at Seoul National University (SNU) since 2021 Sep. Before joining SNU, she was at the civil and environmental engineering, Korea Advanced Institute of Science and Technology (KAIST) from 2014 to 2021. She have the B.S. and M.S. degrees in mechanical engineering from SNU in 2005 and 2007, and the M.S. degree in electrical engineering and the Ph.D. degree in mechanical engineering from the University of Michigan (UM), Ann Arbor, in 2011 and 2012. 

Title: Learning for Decision Making in Cooperative Multi-agent Systems

Abstract: Many real-world problems can be viewed as cooperative multi-agent systems, such as environmental monitoring with teams of aerial vehicles and mixed autonomy traffic. Multi-agent reinforcement learning has shown promise for optimizing decision making polices for such systems. However, a challenge in cooperative multi-agent reinforcement learning is credit assignment, where it can be hard to determine when an agent contributes to a shared reward. In this talk, I will discuss common approaches for addressing this problem and how we use successor features to achieve promising results. I will then discuss our recent work on ad hoc teaming in multi-agent systems, where agents quickly adapt to new teammates.

Bio: Huy T. Tran is an Assistant Professor in the Aerospace Engineering department at the University of Illinois at Urbana-Champaign, Urbana, IL. He received MS and PhD degrees in Aerospace Engineering from the Georgia Institute of Technology, Atlanta, GA, in 2014 and 2015. He then worked at the MITRE Corporation before returning to academia. His research focuses on autonomy in uncertain and unstructured environments, through a combination of reinforcement learning, deep learning, graph theory, and game theory. Current application areas include multi-agent systems, robotics, and intelligent transportation systems.

Title: Planetary Surface Exploration with Heterogeneous Autonomous Vehicles

Abstract: This talk introduces an optimal planetary surface exploration using two heterogeneous agents – the surface rover and the planetary helicopter. The objective of the problem is to maximize the sum of values obtainable by visiting exploration sites. Resource constraints (e.g., fuel and time) associated with the two different agent types are considered. The optimization problem is formulated as a routing class referred to as the vehicle routing problem with profits (VRPP) with two different echelons (two-echelon VRPP). A procedure for obtaining a near-optimal solution to this problem is introduced. A case study with realistic planetary surface exploration parameters demonstrates the effectiveness of the proposed mathematical formulation.

Bio: Jaemyung Ahn received the B.S. and M.S. degrees in aerospace engineering from Seoul National University, Seoul, South Korea, in 1997 and 1999, respectively, and the Ph.D. degree in aeronautics and astronautics from the Massachusetts Institute of Technology, Cambridge, MA, USA, in 2008.

He is an Associate Professor of aerospace engineering with the Korea Advanced Institute of Science and Technology, Daejeon, South Korea. From 2008 to 2010, he worked for Bain & Company as a management consultant helping strategic decisions of clients in various industrial fields. He also worked for the Korea Aerospace Research Institute from 1999 to 2004 and was involved in the research and development of the first liquid propellant rocket and launch vehicle of South Korea as a System Engineer. His research interests include flight dynamics and control and design optimization of complex systems with primary applications in aerospace systems.