Differentiable Motion Manifold Primitives for Reactive Motion Generation under Kinodynamic Constraints

TL;DR

This paper introduces Differentiable Motion Manifold Primitives (DMMP), a neural network-based approach for real-time, reactive motion generation under kinodynamic constraints, combining offline learning with rapid online trajectory search.

Contribution

It extends the Motion Manifold Primitives framework by developing a differentiable, continuous-time neural network architecture trained with offline trajectory optimization data.

Findings

DMMP outperforms prior methods in planning speed.

DMMP achieves higher task success rates.

DMMP ensures better constraint satisfaction.

Abstract

Real-time motion generation -- which is essential for achieving reactive and adaptive behavior -- under kinodynamic constraints for high-dimensional systems is 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 this manifold. Extending the discrete-time Motion Manifold Primitives (MMP) framework, we propose Differentiable Motion Manifold Primitives (DMMP), a novel neural network architecture that encodes and generates continuous-time, differentiable trajectories, trained using data collected offline through trajectory optimizations, with a strategy that ensures constraint satisfaction -- absent in existing methods. Experiments on dynamic throwing with a 7-DoF robot arm demonstrate that DMMP outperforms prior methods in planning speed, task success, and constraint satisfaction.