humanoid.guide introduces KAIST humanoid research prototype

humanoid.guide reported on Nov. 9, 2026, under the headline “humanoid.guide introduces KAIST humanoid research prototype.” We have been unable to verify the information with primary sources, so please refer to the original source for details.

According to humanoid.guide, the robot scores 4/5 for navigation but 0/5 for manipulation, underscoring the need to assess locomotion and object manipulation separately. The report lists a top speed and demonstrations, but provides no test conditions or details of manipulation tasks, so understanding the evaluation method is essential when assessing its applications or comparing it with other robots.

The published scores show a gap between the robot’s locomotion-related rating and its manipulation rating. humanoid.guide gives it an overall score of 4/10, with 4/5 for navigation and 0/5 for manipulation. On this scale, KAIST’s prototype is presented as emphasizing locomotion over the ability to handle objects. However, the definitions of the evaluation categories, scoring method and conditions for comparing it with other robots are not provided, so the scores alone cannot determine the prototype’s overall research level or relative performance. The demonstrations also differ in what they assess. Obstacle avoidance and stair walking show adaptation to the environment, while a 100-meter run demonstrates the ability to travel a set distance. But the report does not state the time taken, the surface or the number of repetitions. The top speed of 12 km/h should also be distinguished from the walking speed of 4 km/h; the top-speed figure does not prove that the robot ran 100 meters at that speed. The available information also does not show how stable the robot was after landing from a jump or what counted as a successful landing. The report says the robot uses in-house-developed quasi-direct-drive motors, motor drivers and a custom-designed 3K compound planetary gearbox, along with reinforcement-learning-based control and onboard vision. This describes a research platform that includes its drive system and control, but the available information does not specify component performance, weight, power consumption, how control tasks are divided or the contents of the training data. Therefore, the results of the locomotion demonstrations cannot be directly attributed to the control approach or the effect of individual components. To assess the robot’s capabilities for research applications, it will be necessary to clarify the scoring criteria and demonstration conditions, including whether the 0/5 manipulation score indicates that the capability has not been implemented or was outside the scope of evaluation. Comparisons will also require the running time and repeatability, the conditions for stairs and obstacles, stability after jumps, and whether manipulation tasks were included. Until such information is provided, the report supports conclusions only about the locomotion demonstrations and the imbalance in its ratings. Please refer to the original source for details.