KAIST’s RAIBO2 ran a marathon on one battery charge and edged past the average human time

# What happened KAIST researchers published a study in Nature describing RAIBO2, a quadruped robot that completed a full 26.2-mile marathon on one battery charge. The run took 4 hours, 19 minutes and 52 seconds. The team focused on reducing energy lost in everyday walking tasks so the robot could finish without recharging and still maintain competitive speed and mobility.

# Why the run matters The accomplishment shows targeted reductions in energy waste can extend operational range for legged robots. RAIBO2 travels about three times farther per charge than comparable quadrupeds and has a projected range near 40 miles. According to the study's lead author, Choongin Lee of KAIST, "What matters is energy consumed per unit of distance." The robot also needs less energy to carry a kilogram one meter than the average human runner.

# How the team cut energy loss The researchers audited where walking robots lose energy and applied hardware and software fixes:

• Motors and drive electronics: About two thirds of a walking robot's energy budget was being lost as heat. The team used low-resistance motor boards and motors that deliver required torque with lower current draw.
• Mass distribution: They cut the weight of each leg and reallocated the saved mass to a larger battery, improving range without increasing overall robot weight.
• Mechanical losses: Friction in gears and foot slip were reduced through design changes and control strategies.
• Regeneration: RAIBO2 can recuperate energy on downhill stretches to recharge the battery while moving.

# Control and training Locomotion was shaped using reinforcement learning in simulation. The reward design encouraged soft foot placement, minimal slipping, and avoidance of motor overheating. Those behaviors lower impact and friction losses, conserving energy in continuous running.

# Limits and remaining work

Adding onboard autonomy will increase power use. Lee and his team are working on smaller onboard neural networks and sensors to improve autonomy while keeping the increase in power consumption minimal. He said that approach aims to add capability without a large rise in energy demand.

# Expert reaction Mechanical engineer Sarah Bergbreiter at Carnegie Mellon University said she was impressed with the balance of efficiency, speed and mobility in RAIBO2. Peers note the work advances endurance for legged robots, which is crucial for practical field uses such as extended inspections or assisted navigation in uneven terrain.

# Bottom line RAIBO2 demonstrates that systematic reductions in hardware and control losses enable legged robots to cover marathon distances on a single charge and outperform average human marathon time by a small margin. The next technical step is adding useful onboard perception and autonomy without eroding the energy efficiency gains.