Overview

Sampling-based techniques are playing an increasingly central role in modern robotics. While gradient-based methods have historically dominated motion generation and control, recent advances in reinforcement learning and sampling-based predictive control have significantly reshaped the landscape. These methods offer compelling advantages in handling complex dynamics, non-smooth objectives, and uncertainty, but they also raise fundamental questions about structure, efficiency, and scalability in robotic decision-making.

This workshop aims to bring together researchers from both theoretical and applied communities to examine the role of sampling in robotic learning, planning, and control. We seek to explore questions such as: 

• How can general-purpose sampling methods be adapted to the specific constraints and structures of robotic systems?

• To what extent should sampling-based approaches exploit the underlying Markov decision process (MDP) structure, and when is it beneficial to relax or bypass this structure altogether? 

• What kinds of theoretical guarantees of convergence and sample-complexity can we provide for these algorithms?

• How can sampling be used to guide, stabilize, or accelerate reinforcement learning and model-based control methods?