ChatGPT Surprises Researchers with Its Spacecraft Piloting Skills

Researchers at an international team have discovered that ChatGPT, a commercially available large language model (LLM), can pilot a spacecraft with remarkable accuracy. In a recent competition, the AI model came in second place, showcasing its potential in autonomous spacecraft navigation.

The Kerbal Space Program Differential Game Challenge, a simulation based on the popular Kerbal Space Program video game, tests the capabilities of autonomous systems in realistic scenarios such as satellite interception and evasion. This competition highlights the growing need for autonomous systems in space exploration, where manual control is impractical due to the vast number of satellites and the limitations of real-time communication over long distances.

To enable ChatGPT to pilot a spacecraft, researchers developed a method to translate the spacecraft's state and goals into text. They then fed this text to the LLM, which provided recommendations on how to orient and maneuver the spacecraft. These recommendations were translated into functional code to operate the simulated vehicle.

The team's approach leveraged the pre-existing training of LLMs on vast amounts of human text, allowing them to achieve the desired results with minimal fine-tuning. This method contrasts with traditional approaches, which require extensive training, feedback, and refinement cycles.

Despite the success, the researchers noted that there is still work to be done, particularly in addressing the issue of 'hallucinations'—unwanted, nonsensical output that could be disastrous in real-world scenarios. However, the results indicate that off-the-shelf LLMs, after digesting extensive human knowledge, can be effectively applied to unexpected tasks.

The competition used an earlier version of ChatGPT, and the researchers are optimistic about the potential improvements with the latest model, version 4. This study underscores the growing role of AI in space exploration and the potential for LLMs to contribute to the development of autonomous systems.

As space exploration continues to expand, the ability to rely on autonomous systems will become increasingly crucial. The success of ChatGPT in this competition opens new avenues for innovation and highlights the potential for AI to play a significant role in the future of space travel.