Hybrid Architectures for Embodied Autonomy: Bridging Learning, Planning, and Control (BLPC 2026)

Our goal is to move the conversation beyond scaling, data, and compute alone. The workshop aims to foster discussion on how learning, planning, and control can be integrated by design, and to highlight methods that turn physical structure into a practical advantage for embodied autonomy.

Embodied autonomy is entering a new scaling era, driven by large datasets, foundation models, and increasingly powerful compute. Yet robots remain physical systems, constrained by dynamics, contact, uncertainty, safety, and limited real-world interaction data. 

This tension raises a central question: 

As learning systems scale, should classical planning and control be replaced, or can they provide the mathematical scaffolding needed to make robot learning more reliable, data-efficient, and deployable?

This workshop is motivated by the latter view. Classical robotics provides structure, while learning-centric methods offer flexibility and adaptability when modeling is difficult or formal assumptions become too restrictive. The key challenge is therefore not whether learning, planning, and control should interact, but how this interaction should be designed.

The central theme of the workshop is this design problem: what should be learned from data, what should remain model-based, and what interfaces should connect learned modules with planning, control, optimization, or verification? By focusing on these questions, the workshop aims to clarify how physical structure can become a practical foundation for modern learning-based autonomy.