As autonomous systems increasingly operate in shared environments alongside humans and other agents, their actions can influence the behavior of those around them. Effective autonomy therefore requires systems to not only understand their surroundings, but also anticipate how other agents may respond to their actions. Developing this capability is fundamental to achieving safe, efficient, and natural behavior in complex, dynamic environments.

 Our research investigates how autonomous systems can incorporate interactions with surrounding agents directly into planning and decision-making. We combine behavior and trajectory prediction with motion planning, optimization, and control to reason about how an agent's actions may influence others and how those responses, in turn, affect the agent's future decisions. This enables autonomous systems to perform complex maneuvers involving cooperation, negotiation, yielding, and adaptation to changing behaviors rather than relying solely on conservative collision avoidance.

A central focus of our work is closing the loop between prediction and planning. Rather than predicting other agents independently and treating their trajectories as fixed obstacles, we develop approaches in which predictions depend on the planned behavior of the autonomous system itself. We explore learning-based models that capture interactive behavior as well as analytical and optimization-based methods that provide robustness, interpretability, and principled solutions. We further investigate scalable approaches that reduce the computational cost of interactive prediction and optimization, helping bridge the gap between sophisticated interaction-aware reasoning and real-time deployment. 

 Ultimately, we aim to enable autonomous systems that can understand interactions, anticipate their consequences, and plan accordingly, allowing them to operate safely, efficiently, and adaptively in complex real-world environments.