SE3-Pose-Nets: Structured Deep Dynamics Models for Visuomotor Planning and Control | Read Paper on Bytez