How RAIBO2 Ran a Marathon on One Battery Charge

RAIBO2’s marathon was an endurance result, not a speed record. Developed at South Korea’s Korea Advanced Institute of Science and Technology (KAIST), the quadruped completed the Sangju Dried Persimmon Marathon alongside human runners on November 17, 2024. It covered 42.195 kilometers in 4 hours, 19 minutes and 52 seconds without recharging or changing its battery. KAIST describes it as the first quadruped to finish a full marathon.3312

How did RAIBO2 last the distance?

The researchers addressed energy loss across the robot rather than relying on a control change alone. RAIBO2 has lighter legs, optimized motor-driving circuits and a machine-learning-enhanced locomotion policy. In combination, those choices reduce the energy needed to move its legs, drive its motors and sustain its gait.1210

Lighter legs require less energy to swing repeatedly. Lower-loss electronics waste less power as heat, while simulation-based reinforcement learning helps develop the robot’s movement policy.1234 The available source excerpts do not quantify how much each change contributed to the marathon result, so the performance is best understood as a whole-system improvement.

What do the energy numbers show?

RAIBO2 consumed approximately 1,280 watt-hours (Wh) over the course—about 30.3 Wh per kilometer. Its reported average cost of transport was 0.25, compared with a cited human value of 0.37. That metric accounts for weight and distance; on that basis, 0.25 is roughly 32% lower. It does not mean the robot was faster than human runners or that battery energy and human metabolic energy have identical upstream costs.12

Reports also describe a range of roughly 65 kilometers per charge, or about three times the per-charge distance attributed to earlier comparable robots. The marathon demonstrated 42.195 kilometers; the additional range was an estimate, not distance completed that day.1013 Battery-capacity figures need care, too: the 2,016 Wh figure associated with the run would imply 736 Wh remaining after 1,280 Wh of use, but another report lists 1,930 Wh of total capacity. That discrepancy makes a precise remaining-charge calculation less certain.32

Why did the race matter beyond the finish line?

The team collected real-world power-consumption and locomotion data during the run, turning the race into a test of long-distance operation rather than only a finish-line demonstration. KAIST’s account also identifies a controller, camera operator and crowd-support staff on the operation team. The available source excerpts do not establish a complete synchronized GPS, electrical and temperature log or verify continuous filming of every part of the course.2433

The findings were published in Nature on September 23, 2026, nearly two years after the race. Reporting described the publication as the first robotics research from a Korean institution to appear in the journal’s main pages.145 KAIST-linked startup Raion Robotics is pursuing applications including inspection and disaster response, where longer operation between charges could be valuable. A marathon finish, however, does not establish performance with equipment payloads, autonomous operation in hazardous settings or the reliability of multi-robot deployments.1921