SCI Autonomy Lab @ UCLA
We study the mathematical science of decision-making in dynamical systems, with a primary focus on safety guarantees and multi-agent decision-making. Core concepts we investigate include nonlinear and hybrid systems theory, reachability analysis, optimal and predictive control, and game theory.
Machine learning provides methods that are generalizable across diverse applications, embodiments, and tasks. Our research focuses on advancing rigorous safety guarantees and principled multi-agent coordination for machine learning-based autonomy.
Building on foundations in control-theoretic analysis and data-driven frameworks, we aim to advance robot intelligence with an emphasis on enabling robots to perform complex tasks that go beyond individual capabilities, tasks made possible only through well-coordinated teamwork.
Innovations in aviation autonomy are enabling new operations such as air taxis and autonomous delivery, while introducing challenges in high-density traffic, uncertain demand, complex environments, and weather. Our research develops robust autonomous air mobility architectures by integrating vehicle-level intelligence with system-level operational decision-making.
Classical control-theoretic analysis provides safety assurance when mathematical assumptions and models hold. We extend these analytical tools to settings where unmodeled effects come into play, such as disturbances, other agents with uncertain intent, and measurement errors. In addition, we study how to effectively balance safety, performance, and learning objectives.
We aim to provide modular safety assurance mechanisms that leverage trajectory data to construct and update safety constraints. These frameworks enable generalizable design methodologies across diverse systems, and building on control theory and statistical learning, deliver provable robustness against model errors, disturbances, and uncertainties.
As individual robotic capabilities advance, the next major challenge is enabling effective coordination among multiple robots and humans. We develop frameworks that support decentralized, coordinated behaviors across many agents, leveraging approaches such as multi-agent reinforcement learning (MARL) and game theory.
Future dense air operations will create tightly coupled interactions across multiple decision-making levels. For example, trajectory planning must account for vehicle-specific health conditions (e.g., battery charge level) and dynamic capabilities (e.g., maximum yaw rate), while simultaneously adapting to fluctuating user demand. We aim to develop methods that deliver system-level safety guarantees for next-generation AAM systems, addressing their multi-level objectives, constraints, and complexities.
We envision teams of mobile manipulators coordinating to transport and assemble objects in a decentralized fashion to achieve large-scale assembly and construction tasks. This vision encompasses a wide range of challenging problems, including task allocation, long-horizon planning, collision-free coordination, and effective communication.
The application is currently closed. We will reopen it in Fall 2026.
The application is currently closed. We will reopen it in Fall 2026.
