A Korean quadrupedal robot has just completed a full marathon on a single battery charge. RAIBO2, developed at the Korea Advanced Institute of Science and Technology, covered 42.195 km in 4 hours 19 minutes, with a transport cost lower than that of a human walker. This result, published in Nature, shifts the debate on mobile robotics: autonomy and energy efficiency constitute a significant barrier to the deployment of quadrupeds, among several material and control constraints.
The essentials
- In outdoor uses where recharging is difficult, greater autonomy can expand employment possibilities, without establishing a general hierarchy of deployment determinants.
- RAIBO2 covered 42.195 km in 4 h 19 min 52 s on a single charge, with a transport cost of 0.25, compared to 0.37 for a human, and more than three times the per-charge autonomy of comparable quadrupeds (Korea Advanced Institute of Science and Technology, Nature, 2026).
- The key mechanism: lightweight mechanics, a motor drive circuit with low resistance, and a locomotion policy designed to limit energy losses.
- The real tension is economic and geographic: these performances open up inspection of linear infrastructure, rescue operations in difficult terrain, and agricultural monitoring, but the unit cost of robots and dependence on Asian supply chains determine the speed of diffusion.
The marathon as a test bench for physical constraints
Quadrupedal robots have long been judged by their ability to overcome obstacles: stairs, rubble, muddy slopes. Boston Dynamics demonstrations fixed this criterion in the collective imagination. But overcoming an obstacle is not the same as sustaining a long-distance mission in a territory where there is no power outlet within reach.
The marathon chosen by the KAIST team as a testing protocol responds to this logic. Forty-two kilometers corresponds to the distance of an inspection round along a medium-length gas pipeline, a sensor survey on an irrigation network, or an initial reconnaissance intervention after a natural disaster. It is an operational specification, not a sporting achievement.
RAIBO2 covered this distance in 4 hours 19 minutes, without recharging. RAIBO2 offers more than three times the distance per charge of existing comparable quadrupeds in the study. The gap changes the class of accessible missions.
“Transport cost” as a measure of locomotor efficiency
The central figure in the study is the transport cost: 0.25 for RAIBO2, compared to 0.37 for a human walking at normal pace. This metric deserves explanation. Transport cost measures the energy expended to move one kilogram of mass one meter of distance. A low number means efficient locomotion. A figure of 0.25 places RAIBO2 below the human benchmark, which was considered a difficult frontier to cross for a four-legged mechanical system.
This result stems from two design choices. First, a motor drive circuit with low resistance to reduce electrical losses. Second, a locomotion policy optimized to limit energy losses, particularly mechanical losses associated with ground contact. Efficient locomotion, in this context, must be planned, not merely reactive.
This point has direct implications for the design of subsequent generations. Efficiency gains result from joint optimization of mechanical, electrical, actuation, and control components, including trajectory optimization through reinforcement learning. It is a constraint of physics, not code.
Inspection, rescue, agriculture: three use cases that become plausible
As long as the autonomy of a quadrupedal robot was measured in a few kilometers, use cases remained confined to bounded industrial spaces: warehouses, construction sites, factories. A human operator could ensure recharging between two rounds. The economic model held within a controlled perimeter.
Beyond thirty to forty kilometers per charge, three sectors change their status.
Inspection of linear infrastructure first. Energy transport networks, pipelines, rural railways span dozens of kilometers between access points. Manual inspection mobilizes teams for entire days. A robot capable of covering this distance without recharging, equipped with thermal or acoustic sensors, reduces the cost and frequency of human involvement in these operations. Countries like South Korea, Japan, and Australia already have automated inspection programs underway for their electrical networks: RAIBO2’s autonomy gives them an argument for extending these programs beyond corridors with easy access.
Rescue in difficult terrain next. After an earthquake or flood, the first hours are critical and traditional vehicles struggle over debris. Enduring quadrupedal robots can map inaccessible zones, locate survivors via audio or thermal sensors, guide human teams. Long-distance autonomy transforms a demonstration tool into an operational tool.
Precision agriculture in difficult environments finally. In mountainous regions or steeply sloped farms, drones poorly cover dense crops and tractors cannot pass. A robot capable of traveling several dozen kilometers per day can ensure phytosanitary monitoring, humidity surveys, early disease detection. Pilot projects exist in South Korea and Japan in this direction; the autonomy constraint was until now the limiting factor.
South Korea and the quadrupedal robot sector in Asia
RAIBO2 is Korean, and this stems from structural choices. South Korea has been investing for several years in mobile robotics as a strategic sector, with KAIST as the central public research hub. The institute has already produced several generations of humanoid and quadrupedal robots, in a dynamic that associates universities, industrial groups such as Hyundai, and public funding agencies.
This concentration is not unrelated to the global industrial geography of robotics. China dominates low-end component production. Japan holds precision actuators. South Korea targets the intermediate segment: high-performance robots, produced at sufficient scale to achieve competitive costs. RAIBO2 fits into this strategy: a performance demonstration that also serves as a showcase for the sector.
On the question of regulation, the Asian ecosystem advances through sectoral experimentation rather than through a general framework. South Korea, Japan, and to a lesser extent China authorize pilot deployments in specific sectors: infrastructure, agriculture, logistics, before generalizing. This is an approach different from the European model, which tends to set the framework before authorizing uses. The speed differences between these two regulatory models have already been documented for humanoid robots.
Persistent limitations of this performance
Robustness in degraded conditions poses a structural design problem. Actuators sensitive to moisture and abrasive particles risk seeing their performance in controlled conditions degraded. Optimizing a component to reduce energy losses tends to restrict tolerances to external stresses, creating tension between locomotor efficiency and resistance to real conditions. This tension concerns material and sealing choices: these choices add weight to the design and mechanically degrade transport cost, and no software solution can substitute for them. Future publications must document to what extent the efficiency gains achieved on stable surfaces are maintained when actuators absorb repeated shocks and significant hygrometric variations.
The KAIST demonstration is clean and documented. It is not enough to predict the speed of diffusion of quadrupedal robots outside controlled spaces.
The first barrier is unit cost. A robot of this class represents a significant investment for operators. For a rural municipality or a medium-sized agricultural operation, the economic calculation depends on local labor costs, maintenance costs, and the actual lifespan of actuators under field stress. Dependence on Asian supply chains also affects diffusion speed. These data are still lacking in the published literature.
The second barrier is robustness in real conditions. The marathon took place over a course with two 50 m climbs. Inspection or rescue missions involve variable surfaces, mud, slopes, debris. Energy efficiency on controlled terrain does not guarantee the same performance on a mountainside after a week of rain. Tests in degraded conditions remain to be documented.
The third barrier, more structural, concerns data infrastructure. An inspection robot produces data: images, measurements, alerts. This data is only valuable if it is processed, stored, and interpreted quickly. In poorly equipped territories—those precisely where the robot could be most useful—connectivity and local processing capacity are often the weak links. The robot’s physical autonomy does not solve the informational autonomy of the system.
These limitations are known to teams working on deployment. They clarify the scope of RAIBO2’s performance without invalidating it. The publication documents an improvement in energy efficiency and autonomy of a quadruped under the tested conditions.
By 2030–2035, if the unit cost of high-autonomy robots continues to fall at the rate observed in other robotics segments, outdoor terrain use cases could become economically accessible for public operators in intermediate-income countries. This scenario remains conditional: it depends on the trajectory of manufacturing costs, robustness in real conditions, and the availability of data infrastructure.
The next bottleneck: embedded energy or mobile recharging
Mobile recharging poses a difficulty that locomotor autonomy alone is not sufficient to solve. Deploying intermediate stations along a route requires prior infrastructure, however light, which contradicts the principle of deployment in unequipped territory. A robot capable of managing its own recharging without fixed infrastructure would need to carry a secondary energy source, but any added mass directly degrades transport cost, partially canceling the gains documented by current tests. Future progress will depend on improvements in energy density and recharging management in open environments.
Beyond locomotor efficiency, future challenges will focus on battery energy density and recharging solutions suited to unequipped territories.
Teams working on field robotics are looking in two directions. The first is wireless or inductive recharging, already tested on agricultural drones in South Korea and Japan: light stations deployed along a route enable continuous rounds without return to base. The second is the use of alternative embedded energy sources, hydrogen cells or integrated solar sensors, that extend autonomy beyond what battery alone can offer.
These developments extend KAIST’s work. RAIBO2 establishes that locomotor efficiency can reach and exceed the human benchmark. The next decisive demonstration will be that of a robot capable of managing its own recharging autonomously, in open terrain, over several consecutive days of mission.
Sources
- Lee et al., “A quadruped robot designed to complete a marathon on a single battery charge”, Nature, 2026, Robotics and Artificial Intelligence Laboratory, Korea Advanced Institute of Science and Technology: https://www.nature.com/articles/s41586-026-11102-5
- PubMed, reference associated with the same publication: https://pubmed.ncbi.nlm.nih.gov/42778604/
