Retour à tous les articles

Offre pour développeurs

Essayez l'API ImaginePro avec 50 crédits gratuits

Créez des visuels propulsés par l'IA avec Midjourney, Flux et plus encore — les crédits gratuits se renouvellent chaque mois.

Commencer l'essai gratuit

AI in Space ChatGPT Shows Unexpected Piloting Skills

2025-07-04Victor Tangermann2 minutes de lecture
Artificial Intelligence
Space Exploration
ChatGPT

A Surprising Co-Pilot

Even though artificial intelligence can be famously baffled by children's puzzles and prone to making things up, new research reveals a surprising capability: piloting a spacecraft. Researchers have discovered that AI models can perform remarkably well when tasked with navigating a simulated vessel through space.

According to a report by Live Science, a collaborative team from MIT and the Universidad Politécnica de Madrid put OpenAI's ChatGPT to the test, tasking it with operating as an autonomous pilot for a pursuing spacecraft. The results were astounding. In a research paper set for publication in the Journal of Advances in Space Research, the scientists detailed how the large language model (LLM) surpassed all expectations. It even secured second place in a space simulation competition that pitted various AI agents against one another, all based on the popular game Kerbal Space Program.

The Kerbal Space Challenge

The testing ground was the Kerbal Space Program Differential Game Challenge, a software design competition created to explore how autonomous agents could manage satellites and other assets in space. The challenge invites participants to use various methods, including reinforcement learning, to have their AI systems face off in different scenarios.

The research team developed a method to translate spacecraft maneuvering instructions into plain text, which they then fed to several commercial LLMs. With minimal prompting, ChatGPT's performance was a standout, earning it the runner-up position in the Game Challenge. It's important to note, however, that first place was awarded to a more traditional system built on established spaceflight equations.

Promising Results and A Reality Check

What makes this even more intriguing is that the study was completed before OpenAI launched its more powerful GPT-4 model in March 2023. This suggests that newer versions of the AI could achieve even more impressive results.

However, we shouldn't get ahead of ourselves and hand over the keys to a real rocket just yet. A major obstacle is that current AI models are still plagued by frequent hallucinations. An AI pilot making up its own flight plan could introduce a terrifying and catastrophic twist to any deep space mission.

More on AI: Tech Workers Say They're Rapidly Being Replaced by AI

Lire l'article original

Comparer les plans et tarifs

Trouvez la formule adaptée à votre charge de travail et débloquez l'accès complet à ImaginePro.

Comparatif des tarifs ImaginePro
PlanTarifPoints clés
Standard$8 / mois
  • 300 crédits mensuels inclus
  • Accès aux modèles Midjourney, Flux et SDXL
  • Droits d'utilisation commerciale
Premium$20 / mois
  • 900 crédits mensuels pour les équipes en croissance
  • Plus de parallélisme et des livraisons plus rapides
  • Support prioritaire via Slack ou Telegram

Besoin de conditions personnalisées ? Parlons-en pour ajuster crédits, limites ou déploiements.

Voir tous les détails tarifaires
ImaginePro newsletter

Abonnez-vous à notre newsletter !

Abonnez-vous à notre newsletter pour recevoir les dernières nouvelles et créations.