When you control a robot from Mars, every command has to be right the first time. Radio signals take 3 to 22 minutes to make a one-way trip, which means controllers on Earth plan each move the day before, send the commands, and wait. Rovers travel tens to hundreds of meters per day. The result is slow science, narrow coverage, and an enormous bill for every Martian day, called a sol.
Gabriela Ligeza, a post-doctoral researcher at the European Space Agency, thinks smaller robots with simpler instruments and more onboard autonomy could break that constraint. In a paper published in Frontiers in Space Technologies, Ligeza and colleagues at ETH Zurich, the University of Zurich, and the University of Basel tested a quadrupedal robot called ANYmal equipped with just two instruments: a microscopic imager and a Raman spectrometer mounted on a robotic arm. Their approach: let the robot decide which rocks to visit, rather than waiting for Earth to pick targets one at a time.
The results were roughly two to three times faster than conventional methods. Multi-target missions, where the robot autonomously moved between sites and ran both instruments on each, wrapped in 12 to 23 minutes. A comparable conventional mission, where operators guided the robot to a single target and waited for data between each step, took 41 minutes. The robot correctly identified gypsum, carbonates, basalts, dunite, and anorthosite at the Mars-analog test facility at the University of Basel, demonstrating that a limited instrument payload is sufficient for geological prospecting when paired with the right autonomy software.
The paper does not argue that legged robots will replace Perseverance or Curiosity. It argues a different architecture: smaller, cheaper rovers that can cover more ground in less time, feeding data back to a human team that then decides where to dig deeper. The Raman spectrometer in this experiment was originally built for the ESA-ESRIC Space Resources Challenge, an ESA program aimed at developing technology for extracting and using materials found on the Moon and Mars. That pipeline connection matters: the instrument is not a research prototype but something that went through a competitive development process designed for actual planetary use.
The gap between this experiment and a real mission is real. The Marslabor facility at Basel uses analogue rocks, analogue dust, and analogue lighting. It does not have Martian radiation, thermal cycling, or the kind of dust storms that have ended missions. The paper does not claim otherwise. Legged robots also face known challenges on loose regolith, where legs can slip in material that wheels handle better. ETH Zurich's Robotic Systems Lab has years of experience with ANYmal in industrial environments, but planetary surfaces are a different problem.
What the paper does demonstrate is that the autonomy stack is worth treating as a first-class design constraint, not an afterthought. The researchers explicitly framed their work against the standard model for Martian surface operations, where the combination of communication delay and limited bandwidth forces a single-target, human-supervised approach. Their alternative: onboard target selection and sequencing, with Earth in a review role rather than a piloting role. Whether that architecture survives contact with actual Martian conditions is an open question. But it is a coherent answer to the constraint that has defined Mars rover operations since Sojourner rolled onto Chryse Planitia in 1997.
For now the data is analog, the robot is Earth-based, and the paper is a single result from a Swiss-Dutch team. The history of space robotics has no shortage of lab results that did not survive the trip to orbit. But the underlying problem, that Mars is too far to drive in real time, has only gotten more acute as missions demand more science per dollar. If smaller, faster, semi-autonomous rovers can prospect multiple sites in a single day instead of a single site in a week, the economics of Mars exploration shift. That is worth watching, even from a lab in Basel.
