Abstract
Real-time motion generation -- which is essential for achieving reactive andadaptive behavior -- under kinodynamic constraints for high-dimensional systemsis a crucial yet challenging problem. We address this with a two-step approach:offline learning of a lower-dimensional trajectory manifold of task-relevant,constraint-satisfying trajectories, followed by rapid online search within thismanifold. Extending the discrete-time Motion Manifold Primitives (MMP)framework, we propose Differentiable Motion Manifold Primitives (DMMP), a novelneural network architecture that encodes and generates continuous-time,differentiable trajectories, trained using data collected offline throughtrajectory optimizations, with a strategy that ensures constraint satisfaction-- absent in existing methods. Experiments on dynamic throwing with a 7-DoFrobot arm demonstrate that DMMP outperforms prior methods in planning speed,task success, and constraint satisfaction.