The Schedule

Introduction: Thai Duong, Lydia Kavraki, Caelan Garrett, and Balakumar Sundaralingam

Invited Talk 1: Evangelos Theodorou

Title:  Improvisation and Disruptive Mechanisms in Planning and Optimization - Going beyond MPPI.

Abstract: Over the last 10 years there have been several efforts in industry and academia towards planning and optimization algorithms that are parallelizable and exploit the structure of the planning problem. Among others, MPPI is one of the most popular parallelizable stochastic optimization methods that is simple to code and deploy on robotic and autonomous systems. At this point, MPPI is broadly used for planning and control tasks while there have been several interesting extensions that aim to achieve safety and improve performance.   In this presentation I will wrestle with the question of how can we go beyond MPPI or algorithms similar to MPPI in a fundamental way?  I will draw inspiration from situations where decision-makers, from single humans to organizations, need to improvise to deal with messy and disruptive events and talk about new paradigms in optimization for planning and control.  If time permits, I will talk about deep unfolding and its use in dynamic optimization for large scale distributed decision-making problems. 

Bio: Evangelos Theodorou is an Associate Professor with the School of Aerospace Engineering at Georgia Institute of Technology and is the Director of the Autonomous Control and Decision Systems Laboratory. Prof. Theodorou is an affiliate of the Institute of Robotics and Intelligent Machines (IRIM) and the Center for Machine Learning Research at Georgia Tech. He is also an Amazon Scholar. His academic and industry research interests span the areas of Control Systems, Optimization, Autonomy, Robotics, Large-Scale Systems and Supply Chain Logistics.

Invited Talk 2: Brian Plancher

Title: Optimizing Robotic Systems at All Scales

Abstract: Intelligent field robots are a promising solution to many societal challenges from combating epidemics, to scaling global supply chains, to providing home health care to the elderly. However, today, robots are mostly limited to laboratory settings as the computational intensity of many robotics algorithms prevents their real-time use on edge robotic hardware. In this talk, I will discuss how I am addressing these challenges through algorithm-hardware-software co-design, generating new algorithms and implementations that can run at real-time rates on the edge. Specifically, I will show how the performance of nonlinear model predictive control (MPC) algorithms can be significantly enhanced through a combination of parallelism, approximation, numerical conditioning, and structure exploitation. Through the resulting theoretical and computational advancements, my work has enabled GPU-accelerated whole-body nonlinear MPC for manipulators at kHz rates over long-horizon trajectories, as well as real-time dynamic obstacle avoidance for microcontroller-powered tiny quadrotors. This work sets the stage for a future filled with dynamic, adaptable, and useful robotic systems.

Bio: Brian Plancher is an incoming Assistant Professor of Computer Science at Dartmouth College (starting Fall 2025) where he will continue to lead the Accessible and Accelerated Robotics Lab (A2R Lab). He is currently an Assistant Professor of Computer Science at Barnard College of Columbia University and is a co-chair for the Tiny Machine Learning Open Education Initiative (TinyMLedu) and IEEE-RAS TC on Model Based Optimization for Robotics. His research is focused on optimizing robotic systems at all scales by developing, optimizing, implementing, and evaluating next-generation algorithms and edge computational systems, through algorithm-hardware-software co-design. He also wants to promote a responsible, sustainable, and accessible future for robotics and edge computing.

Invited Talk 3: Morteza Lahijanian

Title: Kinodynamic Sampling-Based Motion Planning with Massive Parallelism

Abstract: Motion planning under kinodynamic constraints is a fundamental problem in robotics. Every autonomous robot must be able to solve this problem quickly in order to operate effectively in unstructured or dynamic environments. Over the past three decades, sampling-based techniques have emerged as powerful tools for motion planning in high-dimensional spaces. However, these methods still struggle to achieve real-time performance, largely due to their inherently serial computation design. 

In this talk, I argue that parallelization is key for achieving fast performance in sampling-based motion planning (SBMP). I introduce a highly parallel kinodynamic SBMP algorithm explicitly designed for many-core architectures such as GPUs. This tree-based algorithm, called Kinodynamic Parallel Accelerated eXpansion (Kino-PAX), is tightly aligned with GPU execution hierarchies: threads are largely independent, workloads are evenly distributed, and the design leverages low-latency resources while minimizing high-latency data transfers and synchronization overhead. This results in a probabilistically complete algorithm with an exceptionally efficient GPU implementation, achieving millisecond-scale planning times and up to 1000× speedups compared to coarse-grained CPU parallelizations of state-of-the-art sequential algorithms.

Bio: Morteza Lahijanian is an assistant professor in the Aerospace Engineering Sciences department, an affiliated faculty at the Computer Science department and Robotics program, and the director of the Assured, Reliable, and Interactive Autonomous (ARIA) Systems group at the University of Colorado Boulder. He received a B.S. in Bioengineering at the University of California, Berkeley and a PhD in Mechanical Engineering at Boston University. He served as a postdoctoral scholar in Computer Science at Rice University. Prior to joining CU Boulder, he was a research scientist in the department of Computer Science at the University of Oxford. His awards include Outstanding Junior Faculty, Ella Mae Lawrence R. Quarles Physical Science Achievement Award, Jack White Engineering Physics Award, NSF GK-12 Fellowship, and Wadham College Research Fellowship. Dr. Lahijanian's research interests span the areas of control theory, stochastic hybrid systems, formal methods, machine learning, and game theory with applications in robotics, particularly, motion planning, strategy synthesis, model checking, and human-robot interaction. His lab develops novel theoretical foundations and computational frameworks to enable reliable and intelligent autonomy. The emphasis is especially on safe autonomy through correct-by-construction algorithmic approaches.

Invited Talk 4: Maxim Likhachev

Title: Towards massively parallelized search-based planning

Abstract: Planning in physical robots is often severely time constrained. At the same time, the requirements on the fidelity of the planning models and their ability to account for more and more relevant factors keep on growing. Unfortunately, CPUs have hit the plateau in their clock speed, making it hard for single-threaded planning algorithms to support these ever-growing requirements while also respecting time constraints. On the other hand, the number of CPU cores on a typical compute platform has grown significantly, a trend that is likely to continue. This calls for the development of planning algorithms that exploit parallelization. I will talk about some of the work that my group has done towards the development of massively parallelized search-based planning algorithms with robotics being a target domain. In particular, a key feature in planning for robotics is that the major chunk of computational effort during planning is spent on computing the outcome of an action and the cost of the resulting edge rather than searching the graph itself. The algorithms I describe exploit this property to harness the multi-threading capability of modern processors. I will present some of these algorithms, describe their theoretical properties, and show their benefits experimentally.

Bio: Maxim Likhachev is Professor of Robotics at Carnegie Mellon University (CMU), directing Search-based Planning Laboratory (SBPL), and Senior Staff Software Engineer at Waymo. His group at CMU researches heuristic search, decision-making and planning algorithms, all with applications to the control of robotic systems including unmanned ground and aerial vehicles, mobile manipulation platforms, humanoids, and multi-robot systems. He has over 200 publications in top journals and conferences on AI and Robotics and numerous awards including an Influential 10-year paper award, multiple best paper awards, being on a team that won 2007 DARPA Urban Challenge and on a team that won the Gold Edison award in 2013, and other awards and honors. Maxim founded RobotWits, a company devoted to developing advanced planning and decision-making technologies for self-driving vehicles and recently acquired by Waymo, and co-founded TravelWits, an online travel tech company with over $150M in annual transactions that brings AI to make travel logistics easier. Finally, Maxim is an executive co-producer of regional Emmy-nominated The Robot Doctor TV series aimed at showing the use of mathematics in Robotics and inspiring high-school students to pursue careers in science and technology.

Invited Talk 5: Fabio Ramos

Title: Diverse Parallel Motion Planning via Probabilistic Inference

Abstract: Much has been said about the need for diversity in robotics. From diverse datasets for training large vision-action models to diverse motion planners that can infer multi-modal trajectories, the word "diversity" has been a common theme in the last few years of robotics research. But how do we define or even measure diversity? In this talk, I will provide a probabilistic interpretation for diversity and show that  tools designed for deep learning such as differentiable programming languages and parallel computation in GPUs can be conveniently utilized for large-scale probabilistic inference that naturally captures the notion of diversity. I will describe a powerful nonparametric inference method that uses both differentiability and parallelism to provide nonparametric posterior approximations for model predictive control, motion planning, and state estimation. Finally, I will define diversity in trajectory planning in terms of a new mathematical tool–signature transforms–and how it can lead to novel planning methods in the future.

Bio: Fabio Ramos is a Professor in robotics and machine learning at the School of Computer Science at the  University of Sydney and a Principal Research Scientist at NVIDIA. He received the BSc and MSc degrees in Mechatronics Engineering at University of Sao Paulo, Brazil, and the PhD degree at the University of Sydney, Australia. His research focuses on statistical machine learning techniques for large-scale Bayesian inference and decision making with applications in robotics, mining, environmental monitoring and healthcare. Between 2008 and 2011 he led the research team that designed the first autonomous open-pit iron mine in the world. He has over 150 peer-review publications and received Best Paper Awards and Student Best Paper Awards at several conferences including International Conference on Intelligent Robots and Systems (IROS), Australasian Conference on Robotics and Automation (ACRA), European Conference on Machine Learning (ECML), and Robotics Science and Systems (RSS). 

Panel Discussion: David Hsu, Maxim Likhachev, Evangelos Theodorou, Brian Plancher, Caelan Garrett