RAFL: Generalizable Sim-to-Real of Soft Robots With Residual Acceleration Field LearningTuong Cao Gia