Research on RAIBO2, a four-legged walking robot developed at the Korea Advanced Institute of Science and Technology (KAIST) that completed a 42.195-kilometer full marathon without a battery swap, has been published in the main journal of Nature. It is the first time robotics research conducted in South Korea has appeared in Nature's main journal. The team measured the robot's power consumption and locomotion data in real time as RAIBO2 became the first machine of its kind to finish a full marathon course in 2024, producing data that validated design principles for improving energy efficiency.
KAIST said on the 24th that the long-distance locomotion technology and marathon results of RAIBO2, developed by a team led by Professor Hwangbo Jemin of the Department of Mechanical Engineering, were published in Nature's main journal. The paper is titled "A quadrupedal robot designed to complete a marathon on a single battery charge."
RAIBO2 is a four-legged walking robot the Hwangbo team developed in-house with long-distance, high-efficiency locomotion as its goal. After completing the first-generation RAIBO in 2021, the team continued research on extending range and improving energy efficiency, and began designing and building RAIBO2 in 2023.
In November 2024, the team entered RAIBO2 as a registered participant in the Sangju Dried Persimmon Marathon. The robot covered the 42.195 kilometers in 4 hours, 19 minutes and 52 seconds. It was the world's first case of a four-legged walking robot finishing a full course at an actual marathon event, and a defining feature was that its battery was neither recharged nor replaced during the run.
Nature focused less on the finishing time itself than on the design principles the team used to extract more efficiency from the same battery. Competition in the robotics industry has intensified around raising the speed and obstacle-clearing ability of four-legged robots and humanoids, but at actual industrial sites, how long a machine can operate on a single charge matters as much as burst performance. The team treated long-distance travel not as a problem of the battery or the motor alone but as a problem of the robot as a whole system. In a four-legged robot, electricity is lost as heat in the motors when the legs move, and friction occurs at the joints. Impact and slippage when the feet strike the ground also consume energy. Raising efficiency in one part alone therefore makes it hard to extend total range meaningfully.
The analysis found that losses in the electrical system accounted for the largest share of the energy lost across the robot. The main factor was copper loss, in which electrical energy escapes as heat from the motor windings. On the mechanical side, joint friction, landing impact and slippage on the ground also added to energy losses. Simply increasing battery capacity in this situation raises the amount of energy that can be stored but makes the robot heavier. That in turn increases the energy needed to support the body and move the legs.
Moving slowly at all costs was not a solution either, because the computer, sensors and communication equipment keep drawing power even when the robot moves slowly. Going slowly lengthens travel time and raises total power consumption, while going too fast increases losses from landing impact and leg motion. RAIBO2 was found to travel the farthest on the same amount of energy at around 3 meters per second.
The Hwangbo team ultimately optimized each element, from the robot's weight and joint structure to the motors and drive circuits, the battery and gait control. It first made the legs lighter to cut mechanical losses and reduced electrical losses in the motors and drive circuits. It also applied reinforcement-learning-based gait control to reduce landing impact, slippage and unnecessary joint current.
The team verified these design principles in an actual marathon. RAIBO2 was registered as a participant and left an official course record, and researchers ran alongside the robot, filming the entire process of more than four hours from start to finish. At the same time, they recorded the robot's voltage, current and temperature along with battery status across the full course, and linked those readings to position, speed and altitude data measured by GPS to analyze energy consumption section by section. RAIBO2 used about 1,280 watt-hours of energy for the entire run. Based on the finishing data, the team calculated that the robot could travel about 65 kilometers if it used the full battery energy it is able to carry.
"Four-legged walking robots have the advantage of being able to move over rough terrain, but at sites such as mountainous or disaster areas where it is hard to secure a charging location, range can limit the scope of their use," Hwangbo said. "Instead of making the battery bigger, designing the mechanical structure, drive system and gait control together is an approach that can be applied to developing walking robots that have to move for long periods without charging, such as for extended patrols or exploration."
The team's next task is reproducing the performance achieved in a single research robot across multiple actual products. Because RAIBO2 is a research machine for which the researchers designed most of the parts themselves and assembled it in a manner close to handwork, commercialization will require lowering production costs while maintaining performance and reliability in a range of conditions, including waterproofing and dust resistance, impact, temperature swings and electromagnetic interference.
To that end, Hwangbo founded Lion Robotics as a KAIST faculty startup in 2023 and is pursuing the commercialization of four-legged walking robots. Lion Robotics sold nine RAIBO2 units last year and is broadening its reach into defense and industrial settings, including a proof-of-concept project with a unit under the Ministry of National Defense. This year it is expanding production capacity with a target of mass-producing 50 four-legged walking robots, while also pursuing certification related to waterproofing, dust resistance and extreme environments as well as research on military applications such as chemical, biological and radiological detection.
