RAIBO2 Finished a Marathon; Its Maker Now Targets Defense Work

RAIBO2’s 2024 full-marathon finish remains a verified endurance milestone, but it is no longer the robot’s latest chapter. A May 2026 profile of Lion Robotics states that the company sold nine RAIBO2 units in 2025, conducted a proof of concept with South Korea’s Ministry of National Defense and planned to produce 50 units in 2026 while developing robots for the defense market.

Those developments do not establish widespread military deployment or autonomous combat use. They show a narrower progression: the machine that proved it could run for hours on one battery is now being developed and evaluated for work in which operating range, rough-terrain mobility and reliability matter more than race speed.

What RAIBO2 accomplished on the course

KAIST’s official race account records that RAIBO2 completed the 22nd Sangju Dried-Persimmon Marathon in South Korea on November 17, 2024, covering 42.195 kilometres in 4 hours, 19 minutes and 52 seconds; it also identifies hills around the 14-kilometre and 28-kilometre points and calls the finish the first by a quadrupedal robot in an official full marathon.

The world-first description is KAIST’s institutional claim, while the more important technical fact is the documented completion itself. RAIBO2 covered the full road course on a single battery charge, among human participants and without breaking the run into separate short demonstrations.

A marathon tests a different combination of properties from a laboratory speed trial. The robot had to absorb repeated foot impacts, maintain stable motion as the gradient and surface changed, and preserve enough energy to operate continuously for more than four hours. Finishing did not prove that it could navigate independently, recognize mission-relevant objects or function in every weather condition.

The later data show how the battery lasted

The team’s March 2025 RAIBO2 preprint records 1,280 Wh of consumption from a 2,016 Wh battery, an average speed of 2.64 metres per second and 286 metres of total elevation change; it calculates a total cost of transport near 0.25, estimates a 67-kilometre range from the measured consumption, finds roughly three times the travel distance per unit of battery energy of the quadrupeds in its comparison, and shows extra mechanical demand uphill alongside energy generation during descents.

The distinction between measured and estimated distance is essential. RAIBO2 physically completed the marathon distance; it did not complete a separate 67-kilometre road test. That larger figure extrapolates the recorded energy use across the battery’s full stated capacity.

Total cost of transport, or TCOT, normalizes energy use against weight and distance. It is useful for comparing machines of different sizes, but it is not a race ranking and does not mean RAIBO2 outperformed a human marathoner in speed, total metabolism or athletic ability. The achievement was maintaining a low enough electrical cost to complete the course without a recharge.

The downhill generation result also needs context. It does not imply that descending restored all the energy spent climbing, because motors, electronics, joints and foot impacts still introduce losses. It shows that the robot could recover part of the mechanical energy available on a descent instead of dissipating all of it.

Efficiency depended on the complete robot

No single regenerative joint explains the finish. RAIBO2 combined lightweight, force-transparent mechanical components with lower-loss motor-driving electronics and a locomotion policy trained through reinforcement learning. Each part addressed a different source of energy loss while operating as one system.

Quadrupeds must continually apply joint torque to support their bodies, while every foot placement can dissipate kinetic energy at impact. Reducing one loss can worsen another: mechanical choices that alter joint loading can also change the inertia and collision behaviour of the moving legs. The relevant engineering problem is therefore the balance among mechanical design, electrical efficiency and control behaviour.

The learned gait helped manage posture, foot speed and ground contact, while the hardware had to survive repeated loading throughout the run. That combination makes the marathon useful as an integrated endurance test. It still leaves separate questions about sensing, route selection, communications, payloads and resistance to water, dust and temperature extremes.

From endurance demonstration to field product

Lion Robotics’ defense focus changes the meaning of the marathon without turning the race into proof of military readiness. Long operating range can reduce interruptions for charging, and legs can reach terrain that constrains wheeled platforms. Neither advantage establishes that a robot is ready for an operational mission.

The disclosed sales and proof of concept indicate movement beyond a university-only demonstration. They do not reveal continuous deployment, the purchasers’ operational uses, performance under standardized environmental certification or the outcome of independent military acceptance testing. The announced production figure is a company plan, not evidence that all of those units have been built or delivered.

Autonomy requires the same caution. A mobile platform may use autonomous driving and terrain-control software while remaining under human supervision for mission decisions. Nothing in the public evidence supports describing RAIBO2 as an independently operating weapon, and its maker’s defense-market direction should not be confused with fielded autonomous combat capability.

What the marathon record represents now

The original result remains straightforward: RAIBO2 completed an official full marathon on one battery charge with a recorded finishing time. The subsequent technical release made the energy claim more auditable by separating measured consumption from estimated maximum range and by documenting how the integrated system handled changing gradients.

By 2026, the commercially relevant question had shifted from whether the quadruped could finish a race to whether its endurance could support useful work outside it. Sales, a defense proof of concept and a stated production plan provide evidence of that transition, while the absence of documented large-scale deployment keeps the conclusion bounded: RAIBO2 has progressed from a marathon demonstrator toward a field product, but it is not yet proven as a widely deployed defense system.

Also read:

• China’s long-term lunar plans now depend on developing its own Starship
• Nvidia Should Be Printing Money Right Now by Selling AI Hardware. Instead, Its Stock Is Falling
• How’s Allbirds Doing After Its Wild AI Pivot? Now It’s Smartbird — and the Sneakers Are Officially History