2025-07-25 –, Main Room 5
In this talk on TrajectoryBundles.jl, we will discuss the implementation of the Trajectory Bundle Method in Julia leveraging DiffEqGPU.jl. This approach to trajectory optimization is derivative-free, and instead leverages GPU-based ODE solvers and convex optimization.
The Trajectory Bundle Method](https://kestrelquantum.github.io/Piccolo.jl/docs/) was developed in the Robotic Exploration Lab at Carnegie Mellon University as part of the doctoral dissertation of Kevin Tracy. This derivative-free trajectory optimization method leverages the ability of computers to massively sample trajectories of dynamical systems and transform that data into a quadratic program at each optimization step. This is ongoing research and a paper is currently under review.
TrajectoryBundles.jl utilizes the rich ecosystem for solving differential equations, in particular, the package DiffEqGPU.jl to solve ODEs in parallel on GPUs massively. We also leverage Convex.jl and Clarabel.jl to solve a QP at each iteration. Given recent announcements from Nvidia -- e.g. cuDSS -- we are very excited to leverage massively parallel computing in even more facets of this approach.
I am a research associate the Robotics Institute at Carnegie Mellon University. I work on quantum optimal control. My free time is spent climbing, running, cycling, reading, and exploring Pittsburgh.