The Korea Advanced Institute of Science and Technology (KAIST) is showing significant progress in the development of its autonomous machines, going far beyond the boundaries of laboratory demonstrations.
The Humanoid v0.9 prototype developed by Korean engineers demonstrates solid capabilities in performing power operations in a dynamic environment. Deviating from traditional scenarios for carrying lightweight components, the platform manages to independently lift loads weighing up to 20 kilograms, as well as move objects weighing up to 35 kilograms by pushing.
Mastering such physical tasks requires the humanoid robot to constantly restructure its center of gravity. Picking up an object in the air requires full load-bearing from the structure, while pushing it along a surface depends on precise work with friction and current traction. To maintain stability at any moment of execution, Humanoid v0.9 analyzes data from its built-in optical system with cameras and instantly changes its posture relative to the surrounding space.
The key to the machine's precise kinetic behavior lies in the use of deep reinforcement learning algorithms. Through this method, artificial intelligence gradually builds and optimizes its motor models, adapting to complex physical laws without the need to manually code each individual reaction.
The new achievement marks a significant leap compared to the previous generation Humanoid v0.7, whose emphasis was mainly on speed movement and attractive movements, such as Michael Jackson's famous stage walk. With the updated version v0.9, the Korean institute places a clear focus on the practical applicability of humanoid platforms in real industrial and logistics tasks. See more in the video.