Alpha Centauri. (Image Credit: European Southern Observatory/Digitized Sky Survey 2; Acknowledgment: Davide De Martin/Mahdi Zamani)
By the end of 2029, the nonprofit Fermi Explorer Mission is aiming to launch an interstellar spacecraft to Alpha Centauri. To do this, it will use a flight path generated by an AI system called Get Physics Done that Physical Superintelligence (PSI) developed. Alpha Centauri lies 4.4 light-years away, and it may take tens of thousands of years to get there. The team behind this project says their goal is to execute a mission using existing technology and a modest budget.
This idea expanded due to earlier, high-profile interstellar proposals not taking off. In 2016, Yuri Milner announced Breakthrough Starshot. This involved using ground-based lasers to accelerate gram-sized probes at a fifth of light speed. They estimated it would take 2 decades to reach Alpha Centauri. Even though they raised $100 million in pledge funding, the spacecraft never launched ten years later. The founder of Fermi Explorer took a different approach: a solar-electric probe that can be constructed and flown in a few years.
The mission is designed around major cost and mass constraints. With backing from private donors, the team is targeting a $15 million cost. The spacecraft must have a minimum of a 1kg payload, including scientific instruments, artistic items, messages from Earth, and a copy of NASA’s Golden Record. Power became an engineering hurdle for this project. It involved figuring out how to generate sufficient energy for propulsion without massive and heavy solar arrays that would otherwise offset performance gains. For approximately one year, the team tried to find a trajectory that met all those requirements. They didn’t succeed.
Fermi co-founder Philip Johnston then went on a podcast hosted by PSI co-founder Alex Wissner-Gross to discuss the problem. PSI offered to prompt this question to its open-source Get Physics Done AI system. It treats each physics question as a multi-step campaign. Problems are decomposed before the system proposes potential solutions using LLMs like versions of Claude and GPT. Afterward, the software decides which simulation it should run. However, it doesn’t accept answers unless it passes checks from external simulators. This ensures the AI’s reasoning is tied to a numerical validation.
It took the AI one week to generate a trajectory that met the mission’s constraints. The tool recombined well-known orbital maneuvers in a configuration the engineers hadn’t tried before. It suggested the spacecraft slow down to allow its solar orbit to shrink and dip closer to the Sun than Mercury. When the probe is nearby, it would turn its engine on. Since it travels faster at perihelion, the burns get extra orbital energy because of the Oberth effect. And being so close to the Sun allows the panels to get hit with four times the sunlight, enabling short, high-power thrusting. Limiting propulsion to these brief, high-efficiency windows mean the design keeps the mass low while staying within the $15-million budget.
Have a story tip? Message me here at element14.