ELMP: Efficient Learning for Motion Planning via Analytical Policy Gradients

2026-06-30

AI Summary

Existing neural motion planners rely heavily on large amounts of expert demonstration data for adaptation to new environments, incurring substantial costs. This work proposes an efficient self-adaptation approach that directly optimizes the policy through a differentiable kinematics layer and self-supervised fine-tuning, thereby circumventing the need for expensive expert trajectory collection. The method explicitly encodes tool geometry using point-cloud inputs and replaces data-driven updates with analytical policy gradients. Evaluated in previously unseen environments, the approach improves task success rates from 57.3% to 89.8% (averaging 84.8%), reduces cold-start latency by two orders of magnitude compared to conventional methods, maintains millisecond-level inference times, and demonstrates empirical effectiveness on a Franka robotic platform.

Abstract

Neural Motion Planners (NMPs) enable fast reactive motion generation, but adapting them to new environments typically requires recollecting large expert datasets, which is computationally prohibitive. We propose ELMP, a framework for data-efficient adaptation via self-supervised fine-tuning. Rather than generating additional expert trajectories with expensive global planners, ELMP directly optimizes the policy through a differentiable kinematic layer using dense collision, target-reaching, and smoothness objectives. This replaces expert data generation with rapid problem sampling, reducing per-sample adaptation cost by roughly two orders of magnitude. To further support robust generalization across changing kinematic chains, we introduce a mechanism to explicitly encode tool geometry via point clouds. Benchmarked against classical and neural baselines, ELMP achieves an 84.8% average success rate with orders-of-magnitude lower cold-start latency than classical methods. In unseen environments, self-supervised fine-tuning improves success rate from 57.3% (zero-shot) to 89.8%, removing the data collection bottleneck. Our approach maintains millisecond-level inference latency and is validated on a physical Franka Emika Panda robot.

Problem

• Neural Motion Planning
• Data Efficiency
• Adaptation
• Expert Data Bottleneck
• Motion Planning

Innovation

• self-supervised fine-tuning
• differentiable kinematic layer
• data-efficient adaptation
• tool geometry encoding
• neural motion planning