ChatGPT Controls a Spaceship in Simulation, Secures Second Place in Spaceflight Competition
Chatbot performs remarkably in virtual space pilot challenge
Artificial intelligence has proven its potential in space navigation. In a recent simulation competition called the Kerbal Space Program Differential Games, AI chatbots including ChatGPT demonstrated the ability to pilot a spacecraft — and even outperformed many traditional control systems.
The competition, held in 2023, was organized in part by the Massachusetts Institute of Technology (MIT). It aimed to test spacecraft maneuvering skills in low Earth orbit using a simulated environment. One of the featured challenges involved chasing a fast-moving, evasive spacecraft — a scenario designed to mimic real orbital pursuit situations.
In this setup, ChatGPT and LLaMA models were assigned the role of “pursuers.” They were given text-based input such as target coordinates and fuel levels. Based on this information, they had to issue appropriate commands to their virtual spacecraft. Interestingly, the bots also explained their reasoning along the way — with instructions like:
“The target is to the right and above. Apply thrust in that direction, but account for inertia.”
Despite the complexity of the task, the chatbots delivered an impressive performance. According to the results, they outperformed most traditional control systems — securing second place overall. The only system to surpass them was a classical algorithm-based system, similar to the one used in SpaceX’s autonomous Dragon capsule.
What makes this achievement significant is the different approach taken by language models. Traditional onboard computers follow strict mathematical formulas and are designed for pre-programmed scenarios. In contrast, language models like ChatGPT make decisions based on probability, context, and observed patterns, allowing for more flexible problem-solving.
This flexibility can be a major advantage in unpredictable situations — something common in real-world space operations. While standard systems may struggle with new or unexpected scenarios, a language model can adapt its strategy in real time.
However, there is a known limitation. These chatbots are still prone to hallucinations — occasionally generating incorrect or illogical instructions. In a high-stakes environment like spaceflight, even a small error can be catastrophic. That’s why experts stress the importance of using such systems under supervised or hybrid conditions, where human oversight or secondary safety mechanisms are in place.
Still, this experiment highlights the growing potential of language-based AI in real-world applications, far beyond customer service or content generation. As AI continues to improve, its role in autonomous navigation, robotics, and mission control systems may become even more prominent.
With space agencies and private companies alike exploring more advanced autonomy for missions to the Moon, Mars, and beyond, these early test results suggest that natural language-based AI systems could become part of future space missions — not just as assistants, but as active decision-makers.