
모델링된 센서모터 궤적을 통해 현실 세계에서 로봇 제어 작업을 학습하는 유망한 방법론을 제시합니다.


coremsg

humanoid-locomotion-next-token-prediction-for-real-world-control-tasks


Humanoid Locomotion: Next Token Prediction for Real-World Control Tasks


title_rewrite


The author proposes a novel approach to humanoid locomotion by treating it as a next token prediction problem, leveraging sensorimotor trajectories. The core thesis is that modeling sensorimotor data with a causal transformer can enable real-world control tasks through generative modeling.


Humanoid-Locomotion-Next-Token-Prediction-Approach


Humanoid Locomotion: Next Token Prediction Approach
