Welcome and Opening Remarks

Workshop Overview, goals and logistics

Lindon Roberts

University of Melbourne

Title: Handling Noise in Model-Based Derivative-Free Optimization

Abstract: Derivative-Free Optimization (DFO) refers to optimization algorithms that do not require any derivative information, typically from the objective. Among the most practically successful types of DFO are model-based methods, which attempt to construct local interpolation models for use in algorithms that mimic successful derivative-based optimization algorithms. In this talk, I will outline some recent theoretical results for handling of noisy objectives in model-based DFO, considering both generic, bounded noise, and stochastic noise with constraints. This is joint work with Nicole Felice and Sara Shashaani (North Carolina State University).

Ian Abraham

University of Sydney

Title: Using Randomness to Combat Randomness: Towards Robots that Successfully Navigate and Interact in an Uncertain World

Abstract: Random sampling methods for robot control have demonstrated immense potential in expanding the capabilities of robotic manipulation and locomotion.  In this talk, I will showcase studies on how this same random sampling can be used to instill robustness in robot behavioral control for manipulation and effectively counteract unobservable environmental uncertainty. With the use of variational inference methods, I show how this robustness can be harnessed to improve the responsiveness of robots in real-time, enabling highly dynamic manipulation with limited sensor-feedback. Last, I overview recent findings on sampling-based hybrid mode composition that draws from hybrid systems theory to extend the capabilities on whole-body robotic control through fast sequencing of diverse control modes.  

Vince Kurtz

DePaul University

Title: Sampling-based Model Predictive Control on GPU: Why, When, and How?

Abstract: The success of massively parallel reinforcement learning, along with the end of Moore's law, has spurred widespread interest in and accelerated development of GPU-based robotics simulation. Leveraging these massively parallel capabilities for sampling-based MPC seems like a natural choice. But is this really a good idea? Why should we bother implementing sampling-based MPC on GPU? When should we use GPU acceleration, versus leaving everything on the CPU? How should we implement these algorithms, and which implementation details are most critical? This talk will address these questions via case studies involving Hydrax, an open-source library for hardware-accelerated sampling-based MPC built on JAX and MJX/MjWarp. Throughout the talk, we will highlight open research questions in this rapidly developing area of study. 

Poster Teaser Talks

Short presentations by poster authors (1-2 minutes per poster).

Coffee Break & Poster Session

Posters will be displayed throughout the entire workshop.

Informal discussions are encouraged during the break and poster session.

Georgia Chalvatzaki

TU Darmstadt

Title: Sampling Robot Motion in Constrained Domains using Stein Variational Inference

Abstract: Robot motion planning is inherently multimodal, yet classical planning methods typically return only a single solution. The ability to generate and maintain multiple candidate solutions is crucial for improving motion robustness and avoiding poor local minima. In robotics, however, sampled motions must satisfy a range of strict constraints, including collision avoidance, joint limits, contact conditions, geometric constraints, and dynamic consistency. These hard requirements make constrained motion sampling substantially more challenging than unconstrained sampling. In this talk, we will discuss two recent works on constrained motion sampling. The first addresses geometric constraints on SE(3), while the second considers more general nonlinear equality and inequality constraints. Together, these works provide practical approaches for generating diverse, feasible robot motions in complex constrained settings.

Massimo Fornasier

TU München

Title: Consensus-Based Optimization

Closing Remarks