Department of Aerospace Engineering

Back to Listing

AE 590: Professor Marco Pavone, "Real-Time and Dependable Spacecraft Motion Planning: Theory and Algorithms"

Event Type
Aerospace Engineering
103 Talbot Lab
Oct 14, 2013   4:00 pm  
Professor Marco Pavone , Stanford University
Kendra Lindsey
Originating Calendar
Aerospace Engineering Seminars

Autonomous spacecraft maneuvering for proximity operations in dynamic and cluttered environments is a key enabler for several future NASA missions, ranging from missions to the Saturnian system to on-orbit satellite servicing. Autonomous planning for spacecraft proximity operations entails the online solution of a NP-hard combinatorial optimization problem with differential and algebraic constraints, possibly uncertain information, and very limited computational capabilities. This appears beyond the current capabilities of traditional approaches, which are mainly geared toward static and uncluttered environments and often involve a ground station in the loop. In this talk I will discuss a novel approach that leverages recent algorithmic advances in the field of robotic motion planning to spacecraft control. Specifically, I will first discuss some of the unique aspects of the "spacecraft motion planning problem." Then, I will present and discuss a novel sampling-based motion planning algorithm called the Fast Marching Tree algorithm (FMT *), which is asymptotically optimal, appears to converge to an optimal solution faster (and sometimes significantly faster) than its state-of-the-art counterparts, and enables the inclusion in the planning process of typical spacecraft constraints. The FMT * algorithm essentially performs a "lazy" dynamic programming recursion on a set of probabilistically-drawn samples to grow a tree of paths, which moves steadily outward in cost-to-come space. As such, this algorithm is conceptually related to the Fast Marching Method for the solution of eikonal equations. The optimality result is proven with a novel analysis technique that provides convergence rate bounds - a first in this field. I will conclude the talk by giving an overview of related research tasks to mature this technology and by discussing a number of open problems in the field of optimal sampling-based motion planning.

link for robots only