KAIST’s RAIBO2 ran a marathon on one charge by cutting energy loss across hardware and software

KAIST's quadruped robot RAIBO2 ran the Sangju Marathon alongside human runners and finished in four hours, 19 minutes, and 52 seconds. The notable point is endurance: RAIBO2 reached the finish having consumed about 66% of its battery, which the team says translates to a travel range roughly three times farther than many existing legged robots.

Legged robots can go where wheels struggle, but they usually pay an energy penalty for stepping, stopping, and stabilizing. Until now many quadrupeds could only travel on the order of 20 kilometers before stopping for charge. RAIBO2's marathon run pushes that envelope and shows a path to longer-duration outdoor missions such as inspection, search-and-rescue, and extended patrols where recharging is impractical.

The KAIST team approached the problem holistically—tackling multiple sources of energy loss rather than optimizing a single component.

• Motors and electronics: These were the largest drains, about two-thirds of total losses. The team reduced electrical resistance in motor drivers, changed the current-sensing module, and altered controller operation to reduce switching losses and heat dissipation.

• Battery capacity: Weight savings allowed installation of a larger battery with roughly 33% more capacity, extending usable range.

• Learning-based control: An improved AI-driven locomotion strategy lowered the chance of slips or trips and tuned gait timing to reduce inefficient motions at speed.

How this compares to earlier attempts

RAIBO2 had previously run a marathon at the Geumsan Insam Festival but ran out of battery at about the 23-mile mark. Analysis showed frequent speed adjustments increased consumption and caused it to stop short of the finish. The new study integrated hardware and software fixes to address those energy leaks simultaneously.

Carnegie Mellon's Sarah Bergbreiter, who was not involved with the work, told Scientific American she was impressed with the team's ability to combine efficiency with speed and mobility in a quadruped. The research was published in Nature by Choongin Lee and colleagues, and KI AIST authors including Hwangbo Jemin described the motivation: outdoor missions require longer-range operation than current legged robots usually deliver.

Practical implications and next steps

The engineering choices here map to concrete trade-offs teams elsewhere can adopt: reduce motor and controller losses, shed rotating and swinging mass, and invest in control strategies that prioritize steady, slip-free motion. That combination can buy battery capacity or allow smaller batteries for the same range.

Remaining questions include performance on more extreme terrain types, payload trade-offs (how carrying equipment affects range), and how these design choices scale to different sizes or mission profiles. Initial demonstrations like this one show endurance gains are achievable without sacrificing mobility or reasonable speed.