Neural Network Based Optimal Control: Resilience to Missed Thrust Events for Long Duration Transfers – ASC- 2019

A few weeks ago I attended the 2019 AAS/AIAA Astrodynamics Specialists Conference (ASC). This was my first major academic conference in grad school and I was there presenting on my research on using neural networks for optimal control with a special focus on their resilience to missed thrust events.

Where Was I?

The Astrodynamics Specialist Conference (ASC) is one of the big technical astrodynamics conferences along with the Space Flight Mechanics Meeting (SFM) and the Guidance, Navigation, and Control Conference (GNC). This years conference was held in Portland ME, and the northeast in late summer is a gorgeous place to be.

What Was I Presenting On?

With the advent of deep space CubeSats like MarCO and the coming NEA Scout, the number of spacecraft in our deep space fleet is set to grow. While these spacecraft will be great for science and exploration they will place an increasing strain on the limited bandwidth of the DSN. To meet this challenge we will either need expansive infrastructure upgrades to the DSN, or major improvements in spacecraft autonomy. This work focuses on seeing if we can use neural networks as controllers on spacecraft to enhance their autonomy in the realm of deep space navigation. This paper also set out to investigate how well they could recover from outside disturbances, like missed thrust events.

Does it work?

You can read more about my methods in the paper itself, but in summary not only can neural networks can be used for autonomous deep space navigation, but they are also surprisingly resilient to missed thrust events. When no MTEs were present, the network could navigate the spacecraft to the target 99% of the time and once multiple MTEs were introduced, this only dropped the success rate to 79%. This work is quite promising and I plan on expanding it to higher fidelity environments and investigate it’s performance when provided with noisy state information.

Want More Gereshes?

If you want to receive new Gereshes blog post directly to your email when they come out, you can sign up for that here!

Don’t want another email? That’s ok, Gereshes also has a twitter account and subreddit!

2 Comments

  1. Helle Gråberg

    Hi,

    I’m having trouble with the webpage for the paper. Would it be possible for you to email me? Thank you in advance for your answer, I look forward to reading your reasearch.

Comments are closed.