Lunar Advancements – Robotics and AI For Autonomous Exploration

AI-Driven Robotics for Autonomous Space Exploration

One day, space exploration might make use of astronauts living permanently on site, as envisioned by the Artemis missions for the Moon, or by Elon Musk for Mars. Still, even with human presence, a lot of the work in space will be done by robots, if nothing else, because they are a lot easier to replace than human astronauts and a lot less vulnerable to toxic air or vacuum, radiation, brutal temperatures, etc.

Ideally, most of the rovers and robots should be able to handle themselves for simple tasks, with humans on Earth or on-site only involved to help them solve specific problems or determine their daily missions.

As AI progresses quickly, including physical AI, a concept now championed by AI leader NVIDIA, this science-fiction vision might already be a reality.

Scientists are taking the first steps in that direction, both in research projects on Earth and with existing rovers on Mars, with two news items related to this topic in the past few days.

The first one was that NASA has deployed AI assistance to guide the Martian rover Perseverance.

The second one is that researchers at the University of Malaga (Spain), the German Research Center for Artificial Intelligence (DFKI), Sorbonne Université (France), as well as the private companies GMV Aerospace and Defence S.A, Magellium, and Space Applications Services are deploying robots in Earth lava tubes that resemble similar structures on the Moon and Mars¹.

Perseverance Rover’s AI-Assisted Autonomous Navigation

NASA’s First AI-Planned Rover Drives on Mars

NASA’s Perseverance Mars rover hit a new scientific milestone as it completed the first drives on another world that were planned by artificial intelligence. Announced recently, the move was done on December 8^th and 10^th, 2025.

The demonstration used generative AI to create waypoints for Perseverance, a complex decision-making task typically performed manually by the mission’s human rover planners.

Source: NASA

This could prove to be a game-changer for Martian exploration. The extreme distance between Earth and Mars (140 million miles / 225 million kilometers) means that light-lag causes a signal lag, which means that every instruction takes 3-22 minutes (depending on orbital positions) to arrive at Mars from Earth, and feedback then takes the same time again.

As NASA scientists are very cautious to avoid getting the multi-billion dollar project stuck in dust or damaged by a rock, this makes any movement a tedious crawl.

Instead, Perseverance did something new for its 1,707 and 1,709 days on the Martian surface, letting the rover decide where to go using AI.

How It Worked

It used generative AI to analyse the high-resolution orbital imagery from the HiRISE (High Resolution Imaging Science Experiment) camera aboard NASA’s Mars Reconnaissance Orbiter and terrain-slope data from digital elevation models.

Combined with data from previous explorations, this allowed the AI to identify terrain features like bedrock, outcrops, hazardous boulder fields, sand ripples, etc.

The AI model used was Claude, provided by Anthropic, which recently made headlines for potentially disrupting the entire SaaS and software industry, causing a mini stock market crash in this sector.

This AI-guided travel helped Perseverance capture images in its two-hour 30-minute autonomous drive along Jezero Crater’s rim.

AI can also be useful in processing the data generated by space probes and reducing the workload of the robot operators.

No doubt this will be extra useful when actual astronauts are near the robot as well, as by then, AI might be more capable.

In addition, a human presence and logistical support will let NASA operators take more risks, as a robot stuck in dust could be freed manually, instead of causing a catastrophic multi-billion-dollar loss and years of research frozen.

Testing AI On Earth’s Lava Tubes

Why Lava Tubes

While AI deployment on Mars is a groundbreaking first, NASA researchers are understandably cautious in risking a unique asset like Perseverance in an AI experiment. For example, no matter how efficient the AI, it would never take the chance of deploying the robot beyond what could be fixed by a human teleoperator in case something goes wrong.

This is why experimenting with terrains analog to what is found in space, but with Earth resources available nearby, is important as well.

The most important possible terrain on the Moon and Mars is lava tubes, which form natural caves that could form natural shelters for the first astronauts to protect them from cosmic radiation. And thanks to these stellar objects’ lower gravity, lava tubes there tend to be larger than they ever could be on Earth.

Lava tubes can naturally have spots that caved in, leading to holes in the ground providing direct access for exploration.

However, no offworld lava tubes have ever been explored, in large part due to the fact that direct control is impaired by the rock blocking any radio signal.

Testing Robots

The European research team used three different robots working together to explore these extreme underground environments autonomously.

Source: ResearchGate

They deployed their test in the volcanic caves/lava tubes of Lanzarote (Canary Islands).

The system works in 4 phases:

1. The robots cooperatively map the area around the lava tunnel entrance (phase 1).
2. Then the sensorized payload cube is dropped into the cave to gather initial measurements, giving the robots an idea of what to expect (phase 2).
3. Then, a scout rover rappels down through the entrance to reach the interior (phase 3).
4. Lastly, the robotic team explores the tunnel in depth and produces detailed 3D maps of its interior (phase 4).

From Earth Analog Tests to Lunar & Martian Missions

In recent years, the Space Robotics Laboratory at the UMA has worked closely with the European Space Agency, developing algorithms that help planetary exploration vehicles (rovers) plan routes and operate more independently.

Combined with Perseverance’s test run of AI-driven movement, this experiment could form the basis of a new space mission, aiming to explore a lava tube for its potential to form future habitats for early colonization efforts on the Moon and Mars.

This could also have important implications in the search for extraterrestrial life.

Investing In Space Robotics

Intuitive Machines

Sending probes to interstellar objects is going to require a strong expertise in building large space probes and making them arrive in the right place intact. For now, this has been mostly the domain of public institutions like NASA, the ESA, and associated universities.

This is changing as we are getting closer to the point where private companies could start sending automated or manned missions to mine asteroids, especially near-Earth objects.

This sort of project will likely be the next step or done in parallel to the return of manned missions to the Moon, planned for the upcoming years.

Founded in 2013 in Houston, Texas, Intuitive Machines is, for now, a very “Moon-focused” company, as indicated by its stock ticker LUNR, and has already been selected for 4 NASA lunar missions, and employs 400+ people.

Source: Intuitive Machines

It was the first commercial company to successfully land and transmit scientific data from the Moon. It also performed the 1st firing of the LOx/LCH4 (liquid oxygen, liquid methane) engine in space.

The company is working on many projects that will form the base of a lunar infrastructure for exploration and settlement.

The first one is the “data transmission service”, with the technology being tested, and ultimately looking to end with a lunar data transmission constellation around the Moon’s orbit.

Source: Intuitive Machines

The second part is the “Infrastructure as a Service”. It should include an LTV capable of autonomous operations, the telecommunication service, and GPS localization services.

Source: Intuitive Machines

The last segment is the delivery of material to the lunar surface. So far, the company has delivered scientific payloads with the Nova-C lander, a 4.3-meter-tall lander (14-feet) able to deliver 130kg of payload to the Moon.

The next step will be with the Nova-D lander, able to deliver 1,500-2,500 kg of material to the Moon. This payload capacity and size will be the one required for delivery of the Lunar Terrain Vehicle (LTV), as well as the 40kW Fission Surface Power nuclear reactor expected to power the Moon base.

Source: Intuitive Machines

The company has landed many valuable contracts with NASA, for example, the Near Space Network contract, with a maximum potential value of $4.82B.

The LTV contract final decision by NASA between the 3 potential suppliers is expected for the end of 2025, and would be worth up to $4.6B as well.

Besides NASA, the company is trying to diversify its client base, having been selected in April 2025 for a grant of up to $10M by the Texas Space Commission. This will support the development of an Earth reentry vehicle and orbital fabrication lab designed to enable microgravity biomanufacturing.

This reentry vehicle will also provide a backup option and reduce risks for the Company’s future lunar sample return missions.

Another project is the development of low-power nuclear stealth satellites for an Air Force research laboratory JETSON contract.

As the company reaches a positive free cash flow point in Q1 2025, and with the lunar telecommunication contract, it is now becoming a lot safer for investors, moving away from a cash-burning startup to an established services provider to the growing space economy.

And it could form the building block of further deep space exploration and utilization of space resources, especially as it becomes a trusted partner of NASA on par with SpaceX (soon to IPO after its merger with xAI) or Rocket Lab (RKLB -6.47%).

(You can read more about Intuitive Machines in our investment report dedicated to the company.)

References:

¹. Raúl Domínguez et al., Cooperative robotic exploration of a planetary skylight surface and lava cave. Science Robotics (2025). DOI:10.1126/scirobotics.adj9699