A team of researchers has successfully trained the ChatGPT AI chatbot to control a spacecraft in a competition—and interestingly, ChatGPT passed the test with flying colors.
According to Live Science, the researchers initially gave this prompt to ChatGPT: “You are an autonomous agent in control of a chaser spacecraft.” Surprisingly, ChatGPT performed very well and secured second place in a spacecraft piloting simulation challenge.
ChatGPT Can Now Pilot a Spacecraft
In the future, the number of satellites will far exceed what humans can manually manage. Additionally, when exploring deep space, the speed-of-light limitation means we cannot directly control spacecraft in real time. For this reason, researchers have long been interested in developing autonomous systems to manage satellites and navigate spacecraft.
So, if we aim to make deeper explorations in space, we must trust robots and autonomous systems. In a paper soon to be published in Advances in Space Research, the researchers proposed using large language models (LLMs) like ChatGPT and LLaMA to control spacecraft.
Traditional approaches to developing autonomous systems require multiple cycles of training, feedback, and optimization. To streamline this process, the researchers decided to use a large language model instead.
The method ChatGPT used to pilot a spacecraft is quite fascinating. The researchers first translated the spacecraft’s current status and mission objective into textual form. This text was then input into the LLM, which was asked to suggest orientation and maneuvering strategies for the spacecraft.
Next, the researchers designed a translation layer that converted the LLM’s textual output into executable commands capable of controlling a simulated spacecraft. Finally, with just a few short commands and minimal fine-tuning, the researchers were able to prepare ChatGPT to complete many of the tests in the challenge.
