Search Algorithms for Robot Learning

Search-based algorithms provide a natural framework for allocating additional computation at test time to improve decision making, a technique known as test-time scaling, and have played a key role in domains such as the game of Go and language models. However, achieving similar gains for general-purpose robot learning remains an open challenge. Unlike language models, robotic systems must reason over continuous state and action spaces, interact with uncertain environments, and make decisions under real-time constraints. These characteristics pose fundamental challenges for conventional search algorithms and their integration with modern robot learning systems. In recent years, a growing body of work has begun combining search algorithms with robot learning to tackle challenges in real-time continuous control, hierarchical and long-horizon reasoning, and test-time data collection and model adaptation. These developments point to the emergence of a new research direction and motivate this workshop. The goal of the workshop is to build a community around this growing area and identify impactful and emerging research directions by bringing together algorithmic ideas—such as search algorithms, the integration of search and learning, and the role of search in test-time scaling—with application domains including mobile robotics, manipulation, and human–robot interaction, alongside software tutorials to lower the barrier of entry and invite broader participation from the robotics community.

We want to build community and identify emerging research directions in:

• Test-time scaling for robot motion or reasoning 

• Monte Carlo Tree Search for robotics

• Compositional generalization 

• Search and learning architectures 

• Steering diffusion models 

• Guiding autoregressive models

• Planning algorithms and planning with learned models 

• Model-based reinforcement learning 

• And more!

 We are accepting posters! Please submit your poster title and info at this link. At the conclusion of the workshop, we will award a "Best Paper" prize.