Adaptive Distance Kicking Strategy for Humanoid Robots Based on Prior Trajectory Guidance and Multi-Task Residual Learning

The data were obtained from experiments conducted in the RoboCup 3D official platform SimSpark, starting in March 2026.  Table 3 presents the keyframe action data as well as the kicking accuracy and time required to kick the ball under the proposed method for both stationary and rolling balls.  Figure 7 shows the training reward curves comparing Baseline A and the hybrid residual control method.  KickBall.mp4 is the demonstration video of the proposed method.Figure 8 shows the training reward curves of the hybrid residual control method.  Figure 9 shows the training reward curves of the proposed method.  Figures 10 and 11 present the kicking distance errors obtained from testing the proposed method.