• Please check the course syllabus

Materials: Slides

Further reading:

• Differential Games I: Introduction. Isaacs. (1954)

Materials: Notes

Further reading:

• A Time-Dependent Hamilton–Jacobi Formulation of Reachable Sets for Continuous Dynamic Games. Mitchell, et al. (2005)
• Hamilton-Jacobi formulation for reach-avoid differential games. Margellos & Lygeros (2009)
• Reach-avoid problems with time-varying dynamics, targets and constraints. Fisac, et al. (2015)

Materials: Notes

Title: Safety Filters for Uncertain Dynamical Systems: Control Theory & Data-driven Approaches

Abstract: Safety is a primary concern when deploying autonomous robots in the real world. Model-based controllers designed to ensure safety constraints often fail due to model uncertainties present in real physical systems. Providing practical safety guarantees for uncertain systems is a significant challenge, which will be the main focus of this talk. In the first part of the talk, I will review three of the most popular methods for implementing safety filters in general nonlinear systems—Hamilton-Jacobi Reachability, Control Barrier Functions, and Model Predictive Control. The theories underlying each of these methods are well-established for systems with good mathematical models, and have been extended to account for the uncertainties of real-world systems. I will discuss the strengths, drawbacks, and connections of each method. In the second part of the talk, I will discuss how data-driven methods can help resolve the challenge. I will provide an overview of various data-driven safety filters developed during my PhD studies. Finally, I will explore remaining open research problems in addressing safety effectively for real-world robot autonomy.

Materials: Recording

Further Reading:

• An Efficient Reachability-Based Framework for Provably Safe Autonomous Navigation in Unknown Environments. Bajcsy, et al. (2019)
• Data-Driven Safety Filters: Hamilton-Jacobi Reachability, Control Barrier Functions, and Predictive Methods for Uncertain Systems. Wabersich, et al. (2023)
• The safety filter: A unified view of safety-critical control in autonomous systems. Hsu, et al. (2023)

Required reading:

• [P1] On Infusing Reachability-Based Safety Assurance within Planning Frameworks for Human-Robot Vehicle Interactions. Leung, et al. (2020)
• [P2] SHARP: Shielding-aware robust planning for safe and efficient human-robot interaction. Hu, et al. (2022)

Further Reading:

• Trajectron++: Dynamically-Feasible Trajectory Forecasting With Heterogeneous Data. Salzmann and Ivanovic, et al. (2020)
• Identifying Driver Interactions via Conditional Behavior Prediction. Tolstaya, et al. (2021)
• MotionLM: Multi-Agent Motion Forecasting as Language Modeling. Seff, et al. (2023)

Materials: Notes

Title: Behavior Prediction as a Nucleus of Modern AV Research

Abstract: Research on behavior prediction, the task of predicting the future motion of agents, has had an outsized impact on multiple aspects of autonomous vehicles (AVs). From direct improvements in online driving performance to deeper connections between AV stack modules to enabling closed-loop training and evaluation in simulation with intelligent reactive agents, behavior prediction has served as a nucleus for much of modern AV research. In this lecture, I will discuss recent advancements along each of these directions, covering modern approaches for behavior prediction, generalization to unseen environments, tighter integrations of AV stack components (towards end-to-end AV architectures), and methods for simulating the behaviors of agents. Finally, I will outline some open research problems in modeling human motion and their potential impacts on downstream driving performance.

Materials: Recording

Required Reading:

• [P1] Prediction-Based Reachability for Collision Avoidance in Autonomous Driving. Li, et al. (2021)
• [P2] Probabilistically Safe Robot Planning with Confidence-Based Human Predictions. Fisac & Bajcsy, et al. (2018)

Title: Inverse games: an MPEC by any other name…

Abstract: This lecture will introduce mathematical programs with equilibrium constraints (MPECs), and show how they encompass an “inverse” variant of mathematical games in which parameters of players’ costs and constraints must be inferred from data, or designed to yield specific equilibrium outcomes. We will begin with a brief review of the fundamentals of constrained optimization, and discuss how these familiar concepts appear in inverse games and the implications for designing efficient solution methods. The lecture will conclude with a review of several recent papers that present new developments in this space.

Materials: Recording

Further Reading:

• Quantifying Hypothesis Space Misspecification in Learning from Human-Robot Demonstrations and Physical Corrections. Bobu et al. (2020)
• Confidence-Aware Imitation Learning from Demonstrations with Varying Optimality. Zhang and Cao. (2021)

Required Reading:

• [P1] Deep reinforcement learning from human preferences. Christiano et al. (2017)
• [P2] Contrastive Preference Learning: Learning from Human Feedback without RL. Hejna et al. (2023)

Further Reading:

• Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback. Capser et al. (2023)
• Nash Learning from Human Feedback. Munos et al. (2024)

Required Reading:

• [P1] Learning Autonomous Vehicle Safety Concepts from Demonstrations. Leung, et al. (2023)
• [P2] Learning Shared Safety Constraints from Multi-task Demonstrations. Kim & Swamy, et al. (2023)

Required reading:

• [P1] LS3: Latent Space Safe Sets for Long-Horizon Visuomotor Control of Sparse Reward Iterative Tasks. Wilcox & Balakrishna, et al. (2021)
• [P2] Discovering Closed-Loop Failures of Vision-Based Controllers via Reachability Analysis. Chakraborty & Bansal (2023).

Presenters: Bowen Jiang, Yilin Wu, Weihao (Zack) Zeng, Samuel Li, Sidney Nimako – Boateng, Xilun Zhang

Presenters: Jehan Yang, Eliot Xing, Yumeng Xiu, Kavya Puthuveetil