Legged robot performing analogue tests in Marslabor at the University of Basel. (Credit: Dr Tomaso Bontognali)
Mars Exploration Has A Speed Problem. This Bot May Have Solved It.
In A Nutshell
- Researchers tested a four-legged robot called ANYmal in simulated Mars and lunar environments, having it autonomously survey and identify rock samples without real-time human guidance.
- In its best run, the robot correctly identified all three geological targets in 15 minutes, roughly twice the data collection rate of the human-supervised lunar mission.
- The robot flagged gypsum and carbonate minerals, both detected on real Mars and considered strong candidates for preserving evidence of ancient life.
- Precise arm placement was the biggest weakness, with one carbonate rock cluster missed in three of four Mars missions, pointing to targeting accuracy as the key challenge to solve next.
Finding a promising rock on Mars often takes a full Martian day or more. Scientists on Earth analyze the previous day’s data, decide what to investigate, and send up commands across a radio gap that can stretch to 22 minutes one way. By the time the rover responds, another planning cycle has already begun. Researchers at the University of Basel and ETH Zurich think there is a faster way: a four-legged robot that doesn’t need step-by-step instructions from Earth.
In its best run of several simulated planetary missions, a semi-autonomous robot called ANYmal surveyed three geological targets, identified each one’s mineral composition using two onboard instruments, and wrapped up in 15 minutes, correctly identifying all three. Among the rocks it flagged were gypsum and carbonate, two minerals scientists consider top candidates for preserving evidence of ancient Martian life. Both have been detected on Mars, and both can lock organic molecules inside their crystal structures. A robot capable of finding them without waiting for step-by-step guidance from a team millions of miles away could give future missions a meaningful head start.
The research, published in Frontiers in Space Technologies, pitted semi-autonomous, multi-target surveying against traditional human-supervised operations in side-by-side simulated trials, measuring how the two approaches compared in speed, accuracy, and scientific value.
Why Autonomous Mars Robots Can’t Afford to Wait
The Curiosity and Perseverance rovers are remarkable machines, but their scientific operations are tightly managed from Earth. Mission teams spend each Martian morning reviewing the previous day’s data, setting new priorities, and sending up a fresh command sequence. With one-way signal delays ranging from 3 to 22 minutes depending on the planets’ positions, a single back-and-forth between rover and mission control can eat up nearly an hour. Rovers can navigate and avoid obstacles on their own, but neither conducts a full scientific survey from start to finish without human input mid-run.
Robots pre-loaded with science objectives and trusted to execute them independently sidestep that problem. As missions push farther from Earth, it only gets more pressing. NASA’s Dragonfly rotorcraft, bound for Saturn’s moon Titan, will face one-way signal delays exceeding an hour, making real-time ground control functionally impossible.
A Dog-Sized Robot With a Science Lab on Its Arm
ANYmal, made by Swiss company ANYbotics, is a four-legged robot about the size of a large dog, weighing about 130 pounds (60 kg). Researchers fitted it with a custom robotic arm carrying two instruments: MICRO, a close-up imager that photographs rock surfaces under visible, ultraviolet, and infrared light to reveal texture and structure, and a Raman spectrometer, a laser-based tool that identifies mineral composition by measuring how molecules scatter light, with no contact required.
Testing took place at the Marslabor, a University of Basel facility with a roughly 430-square-foot Mars-analog test bed built to support preparation for missions including ESA’s ExoMars. Mars runs used rocks chosen for their scientific relevance, including gypsum, carbonate, sandstone, and sulfur-coated basalt, all under lighting calibrated to mimic the Martian surface. Lunar runs were conducted in near-darkness, lit by a single low-angle floodlight simulating the harsh shadows near the Moon’s south pole.
15 Minutes, Three Targets, One Standout Mission
Two approaches were compared across five mission runs. In four semi-autonomous Mars missions, operators selected targets upfront using camera images and a terrain map, then handed control to the robot. ANYmal walked to each rock, deployed its instruments, recorded the data, and moved on without further input. In one human-supervised lunar mission, an operator directed each step, reviewing results before deciding what came next.
Results were mixed but instructive. Three of the four Mars missions correctly identified two out of three targets, a 67 percent success rate. Mission 3 was the standout, hitting all three targets in 15 minutes. The recurring weak point across all four Mars runs was a cluster of carbonate rocks that the arm successfully reached in only one of four attempts, landing on surrounding soil in the other three. Still, “the success of Mission 3 highlights the potential of automated, multi-target sampling strategies for rapid planetary exploration,” the authors write.
Mars missions gathered data roughly twice as fast as the human-supervised lunar run, though with less opportunity to refine measurements. The lunar mission took 41 minutes but generated nearly twice as much useful scientific data overall, benefiting from an operator’s ability to adjust in real time.
Gypsum was correctly identified in all four Mars missions. Carbonate was flagged in Mission 3. Sulfur-bearing basalt, which can contain minerals linked to past water activity and are considered promising in biosignature searches, was confirmed in every run. When close-up images came out blurry from arm vibration, the Raman spectrometer produced clean chemical readings regardless.
What an Autonomous Mars Robot Could Mean for Finding Ancient Life
Precise arm placement was the system’s biggest weakness, and the carbonate cluster results make that plain. Researchers flag visual feedback systems that let the arm self-correct mid-deployment as the most important near-term fix.
The spectrometer’s detection range presented a separate limitation: it cannot detect water ice, which matters considerably for lunar south pole missions where ice is a primary target. A broader-range instrument would be needed for that work.
Mars exploration moves in deliberate, slow cycles dictated by the speed of light and the caution of mission planners. A robot that can survey a site, identify the most scientifically promising rocks, and deliver results before the next planning session could, in principle, compress multiple planning cycles into a much shorter window, and do it without anyone on Earth lifting a finger.
Paper Notes
Limitations
This study was conducted entirely in a controlled laboratory setting rather than in the field, which limits direct comparison to actual planetary surface conditions. All experiments were run in Earth gravity; lower gravity on the Moon or Mars would change the robot’s locomotion and arm dynamics, though the researchers note reduced gravity would likely make arm manipulation easier rather than harder. Rock samples tested were small, ranging up to about 30 centimeters in diameter, and the mineral set was limited. The Raman spectrometer operated at 785 nanometers with a detection range of 400 to 2,300 inverse centimeters, excluding water ice detection, a meaningful limitation for lunar surface applications where ice identification is a primary science goal. Sample contamination affected at least one result: residual sulfur powder on the microscopic imager’s contact foam appears to have transferred to a dunite rock sample, producing a misleading Raman signal. The data usefulness scoring system is described by the authors as a semi-quantitative heuristic, meaning scores reflect ordered judgments rather than precise measurements. Only one lunar mission cycle was conducted, compared to four Mars cycles, limiting direct statistical comparison between the two operational approaches.
Funding and Disclosures
This research was supported by the Swiss National Science Foundation (grant 20021_197293) and by the European Space Agency and European Space Resources Innovation Centre through ESA Contract No. 4000137333/22/NL/AT and ESA Contract No. 4000141520/23/NL/AT. The Raman spectrometer used in the study was provided by Metrohm Schweiz AG, which is acknowledged in the paper. Authors declared no conflicts of interest and stated that generative AI was not used in the creation of the manuscript.
Publication Details
The study, “Semi-autonomous exploration of martian and lunar analogues with a legged robot using a Raman-equipped robotic arm and microscopic imager,” was authored by Gabriela Ligeza, Philip Arm, Tomaso R. R. Bontognali, Valentin T. Bickel, Hendrik Kolvenbach, Nikolaus J. Kuhn, and Florian Kehl, representing the University of Basel, ESA-ESTEC, ETH Zurich, the Space Exploration Institute in Neuchâtel, the University of Bern, and the University of Zurich. Published March 31, 2026, in Frontiers in Space Technologies. DOI: 10.3389/frspt.2026.1741757.
