Year in Review for the MoSAIC Center

It has been an exciting year since the 2022 launch of MoSAIC. Our team has expanded the capacity of the Institute of Transportation Studies' (ITS) UC Davis technical capacity on connected and automated vehicles (AVs), artificial intelligence (AI), and robotics. These efforts are in support of MoSAIC’s vision to develop a science of safe, sustainable, and inclusive mobility to transform transportation in ways that reduce greenhouse gas emissions, traffic congestion, cost, and transportation inequities.

We are taking steps to contribute to the new field of mobility science, which hinges on leveraging multi-disciplinary expertise in AI, social and behavioral science, AV technology, policy, environmental science, and healthcare. To realize our vision, we have conductef research, workforce development, and knowledge transfer, in ways that fully incorporate diversity, equity and inclusion (DEI). MoSAIC produced several publications over it's first year, some highlighted below.

2023 Publications

Improving AV Safety by Contributing to the Advancement of AI Foundations. This research aims to advance the foundation of AV operations by looking for opportunities to improve the outcomes of deep learning algorithms. A promising approach is warm-start strategies, where both offline and online reinforcement learning can converge more efficiently. MosAIC researchers have developed reinforcement learning-based behavior/trajectory planning and real-time control algorithms for AV driving while reducing the training time significantly.

1. Warm-Start Actor-Critic: From Approximation Error to Sub-optimality Gap. International Conference on Machine Learning 2023 (`notably top 5% accepted papers among 6500+ submissions).
2. Continual Learning of Generative Models with Limited Data: From Wasserstein-1 Barycenter to Adaptive Coalescence. IEEE Transactions on Neural Networks and Learning Systems.
3. CLARE: Conservative Model-Based Reward Learning for Offline Inverse Reinforcement Learning.' Proc. International Conference on Learning Representations (ICLR).
4. Communication-Efficient Distributed Learning: An Overview. IEEE J. Sel. Areas Commun.

Improving AV Societal Outcomes with Effective Policy. This research area aims to identify solutions for regulatory safety frameworks. Recent projects conclude that there is more to do to improve safety outcomes and encourage accountability of innovators in this maturing industry. We met with cities, community groups and industry partners on the front lines of U.S. deployment. We also evaluated gaps in state and federal efforts and made suggestions on how to ensure baseline safety and continuous improvement.

1. D’Agostino, M.C.; Michael, C.E; Ramos, M. “Automated Vehicle Safety Policy Analysis: Identifying Risks and Developing Effective Frameworks” UC Institute of Transportation Studies. Oct 2023 (review draft – inquire for access)
2. D'Agostino, M.C.; Michael, C.E; Venkataram, P. “Learnings from Early Automated Vehicles Deployments in U.S. Cities” UC Institute of Transportation Studies. Oct 2023 (review draft- review draft – inquire for access)

• MoSAIC research was highlighted by Professor Junshan Zhang at the International Conference on Machine Learning (ICML) 2023 in Honolulu, HI highlighting his work on Warm-start reinforcement learning, which has been successfully applied in AlphaZero and ChatGPT, demonstrating its great potential to speed up online learning. MoSAIC findings reveal that a “good” warm-start policy (obtained by offline training) may be insufficient, and bias reduction in online learning also plays an essential role to lower the suboptimality gap.
• The article Don’t fall prey to the current panic over automated vehicles, was published by The Hill, September 17, 2023, and authored by Dan Sperling & Mollie Cohen D'agostino. The article highlights the need to contextualize stop and go issues occurring in early AV deployments, and it encourages better policy to improve outcomes on data transparency, pooling, and disability access.
• Mollie Cohen D'agostino was a featured speaker at the International Workshop on Autonomous Systems Safety, Southampton, UK, where she highlighted the role of policy research in determining the best path forward for autonomous automotive safety and risk analysis.
• The participation in this event builds on one of the event's co-sponsors, the partnership between MoSAIC and the B. John Garrick Institute for the Risk Sciences at UCLA. 
• MoSAIC provided expert testimony at a recent California Public Utility Commission Public Workshop on Autonomous Vehicle Data Reporting held on June 22, 2023. 
• MoSAIC organized three meetings with Assembly member Aguilar-Curry regarding the now failed California Assembly Bill 316, which would have required safety drivers in all medium- and heavy-duty AVs. These meetings provided context for the AV freight landscape and connected Assembly staff to researchers focusing on risk analysis.  

Several introductory research projects are underway with support from partners at the UC Institute of Transportation Studies and the State of California, and federal funding.

Current MoSAIC projects include:

• Automated vehicle trucks: a study of labor and policy considerations
• Safe vehicle headway for AVs – a traffic analysis study
• Equitable AV demonstration incubator seed project
• Testbed for convergence of AI, sensing, communications & control
• Inverse reinforcement learning (IRL)
• Conservative reward learning for offline IRL
• Joint planning and vehicle control
• Warm-start reinforcement learning aid by offline training