NASA’s Jet Propulsion Laboratory and university partners are building a case for sending four-legged robots, modeled on the agility of dogs, to explore the Moon and Mars with minimal human oversight. The concept targets terrain that wheeled rovers cannot handle: steep crater walls, boulder fields, and underground lava tubes where radio signals drop out entirely. If the technology matures on schedule, these machines could become the first autonomous scouts to map subsurface caves on another world.
Why Legs Beat Wheels on Other Worlds
Wheeled rovers have defined planetary exploration for decades, but their design limits where they can go. Rocky slopes, loose regolith, and cave entrances present obstacles that a wheeled platform either avoids or cannot cross at all. Legged robots sidestep those constraints. A quadruped can step over rocks, recover from stumbles, and crouch through narrow openings, much like a biological animal navigating uneven ground.
NASA recognized this advantage early. A technical concept archived on the NASA reports server proposed a small dog-like quadruped robot powered with McKibben air muscles for Mars-like terrains. That work, dating to the mid-2000s, laid out the basic argument for legs over wheels: greater adaptability on rough surfaces and the ability to handle elevation changes that would strand a conventional rover.
The idea has since evolved from pneumatic prototypes to sophisticated platforms equipped with lidar, stereo cameras, and onboard AI. What has not changed is the core engineering bet: legs offer a path to places no wheeled machine can reach, and the scientific payoff of those places, particularly subsurface caves that may shelter water ice or biosignatures, justifies the added complexity.
From DARPA Tunnels to Lunar Caves
The strongest real-world proof came from the DARPA Subterranean Challenge, where JPL’s Team CoSTAR deployed Boston Dynamics’ Spot robot fitted with custom autonomy software. In underground courses designed to simulate mines, tunnels, and cave networks, the robots had to navigate underground while coping with frequent communications dropouts. The scenario closely mirrors what a robot would face inside a lunar or Martian lava tube, where line-of-sight radio contact with a surface base station disappears within meters of the entrance.
JPL researchers Ali Agha and Joel Burdick, quoted in the agency’s own reporting on the challenge, stressed that these robots “have to explore on their own.” That phrase captures the operational reality of deep-space robotics: a command signal from Earth to Mars takes between four and 24 minutes one way, and inside a cave, even a nearby human operator may lose contact entirely. The robot must decide where to step, how to map its surroundings, and when to turn back, all without waiting for instructions.
The autonomy framework behind those decisions is documented in a system-level paper describing multi-robot exploration under severe communications constraints. The NeBula architecture, developed by Team CoSTAR, incorporates risk-aware planning and decentralized reasoning so that each robot in a group can act independently when the network fragments. That capability matters beyond caves: any mission on the far side of the Moon or in a Martian canyon will face intermittent blackouts.
Autonomy Stacks Designed for Extreme Terrain
Giving a legged robot the physical ability to walk is only half the problem. The other half is building software that lets it decide where to walk. A peer-reviewed paper presented at IEEE/RSJ IROS 2020 described an autonomy stack for long-range exploration in extreme environments using Spot. The system demonstrated exploration behaviors in GPS-denied settings, conditions directly analogous to the Moon and Mars, where no satellite navigation exists.
More recently, researchers published the LASSIE framework, short for Legged Autonomous Surface Science In Analogue Environments. The paper, available on arXiv, describes a verified approach tested through lab work and field deployments in planetary analog terrain. LASSIE integrates supervised autonomy with environment-aware planning, allowing a legged robot to make scientifically productive decisions about where to step and what to sample while a human supervisor sets high-level goals rather than issuing step-by-step commands.
This split between high-level human direction and low-level robotic autonomy is the practical meaning of “limited human control” in the headline. No one is proposing to send a robot dog to Mars and forget about it. The model is closer to a field biologist directing a trained search dog: the human chooses the search area, the dog decides how to cover it.
NASA’s Broader Push Toward Autonomous Teams
Legged robots are part of a wider agency effort to reduce the need for constant human oversight on other worlds. NASA’s CADRE project, a network of small Moon-bound rovers designed to work together without explicit commands, demonstrates the same principle with wheeled platforms. The rovers coordinate their movements and divide tasks among themselves, relying on onboard logic rather than Earth-based operators to resolve conflicts or reroute around obstacles.
CADRE uses wheels, not legs, but the autonomy architecture shares DNA with the legged-robot programs. Both require decentralized decision-making, tolerance for communications delays, and the ability to map unknown environments in real time. The difference is terrain access. Where CADRE’s small rovers are designed for relatively flat lunar surfaces, quadrupeds aim for the places those rovers cannot go.
NASA has also funded university teams to prototype alternative locomotion for extreme lunar terrain. Through the BIG Idea Challenge, student teams developed concepts including an articulated rover that can reconfigure its body to climb steep slopes and traverse loose regolith. A detailed summary on the technical reports server describes how these student-built systems are evaluated in analog environments, emphasizing robustness on slopes, power efficiency, and the ability to recover from partial failures.
Legged robots fit naturally into this portfolio. Their ability to adjust individual limb trajectories, brace against rocks, and redistribute weight when one foot slips gives them a built-in resilience that rigid-bodied rovers lack. In a future mission architecture, NASA could deploy a mixed team: small wheeled scouts for broad surface mapping and heavier quadrupeds for targeted excursions into hazardous zones.
Designing for Limited Human Control
Operating with limited human control is not just about autonomy software; it shapes every aspect of mission design. Power management, for example, must be conservative enough that a robot can retreat from a cave before its batteries run low, even if communications are lost. LASSIE-style planners explicitly encode safety margins so that a legged robot never commits to a path it cannot reverse within its remaining energy budget.
Communication strategies also change when robots are expected to fend for themselves. The NeBula framework envisions robots dropping small radio relays or using other members of the team as mobile communication hubs. If a quadruped descends into a lava tube, a second robot might wait near the entrance, relaying intermittent status updates to an orbiter or surface lander whenever line of sight is restored.
Science operations adapt as well. Rather than uplinking a detailed daily command sequence, mission controllers might send a set of priorities: map this side passage, search for temperature anomalies, or collect high-resolution images of unusual rock formations. The robot’s onboard software would then choose specific waypoints and routes that satisfy those goals while respecting its own mobility limits.
From Analogue Sites to Real Missions
For now, most of this work happens on Earth. JPL and its partners test quadrupeds in mines, lava tubes, and volcanic fields that mimic the low-light, cluttered conditions expected on the Moon and Mars. These analogue campaigns are more than dress rehearsals; they expose the autonomy stacks to real dust, uneven footing, and sensor noise that cannot be fully replicated in simulation.
Each field season feeds back into the software. Failures, such as a robot getting stuck on loose gravel or misclassifying a shadow as a solid obstacle, translate into new training data and algorithm updates. Over time, the goal is to build a level of robustness where legged robots can be trusted to operate for hours without human intervention, even when conditions deviate from expectations.
Hardware is evolving in parallel. Commercial quadrupeds like Spot provide a convenient testbed, but a flight-ready planetary robot will need radiation-hardened electronics, dust-resistant joints, and thermal control systems tuned to the brutal temperature swings of airless worlds. Engineers are exploring lightweight leg designs, modular payload bays for science instruments, and self-righting mechanisms that allow a toppled robot to stand up without help.
Scouting the Hidden Frontiers of the Solar System
If these elements come together (rugged hardware, risk-aware autonomy, and team-based mission architectures), legged robots could open a new frontier in planetary science. Lava tubes on the Moon may preserve ancient volcanic deposits and offer natural shielding from radiation, making them prime candidates for future human habitats. On Mars, caves could trap volatiles like water ice and protect organic molecules from surface chemistry and cosmic rays.
Sending a quadruped into such environments would be a high-stakes experiment in trust. Mission teams would relinquish the comfort of minute-by-minute control in exchange for access to places that have remained hidden since their formation. The payoff could be transformative: maps of underground voids, temperature and humidity profiles of sheltered niches, and close-up imagery of rock surfaces untouched by the planet’s harsh surface environment.
In that sense, robot dogs on the Moon or Mars are not a gimmick but a logical extension of decades of work in autonomy, mobility, and multi-robot coordination. By embracing limited human control, NASA and its partners are betting that the next great discoveries may lie where radio contact fades and only a sure-footed machine can go.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
