# ELMP: Efficient Learning for Motion Planning via Analytical Policy Gradients

arXiv: https://arxiv.org/abs/2607.00215 · HTML: https://mlsanity.com/paper/2607.00215/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in RL with diffusion and flow models, one of 15 directions in Deep reinforcement learning methods.
- Builds on 21 of its 36 references that are in the map.
- First posted on arXiv 30 Jun 2026.

## Placed in

- Deep reinforcement learning methods / [RL with diffusion and flow models](https://mlsanity.com/deeprl/rl-with-diffusion-and-flow-models/)

## Builds on (most cited first, 21 of 21)

- [PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space](https://mlsanity.com/paper/1706.02413/) (2017)
- [Learning Quadrupedal Locomotion over Challenging Terrain](https://mlsanity.com/paper/2010.11251/) (2020)
- [Brax -- A Differentiable Physics Engine for Large Scale Rigid Body Simulation](https://mlsanity.com/paper/2106.13281/) (2021)
- [Learning by Cheating](https://mlsanity.com/paper/1912.12294/) (2019)
- [Accelerated Policy Learning with Parallel Differentiable Simulation](https://mlsanity.com/paper/2204.07137/) (2022)
- [Gradients are Not All You Need](https://mlsanity.com/paper/2111.05803/) (2021)
- [Motion Planning Networks: Bridging the Gap Between Learning-based and Classical Motion Planners](https://mlsanity.com/paper/1907.06013/) (2019)
- [Motion Policy Networks](https://mlsanity.com/paper/2210.12209/) (2022)
- [STORM: An Integrated Framework for Fast Joint-Space Model-Predictive Control for Reactive Manipulation](https://mlsanity.com/paper/2104.13542/) (2021)
- [Learning with 3D rotations, a hitchhiker's guide to SO(3)](https://mlsanity.com/paper/2404.11735/) (2024)
- [A Smooth Representation of Belief over SO(3) for Deep Rotation Learning with Uncertainty](https://mlsanity.com/paper/2006.01031/) (2020)
- [Batch Informed Trees (BIT*): Informed Asymptotically Optimal Anytime Search](https://mlsanity.com/paper/1707.01888/) (2017)
- [Training Efficient Controllers via Analytic Policy Gradient](https://mlsanity.com/paper/2209.13052/) (2022)
- [Adaptively Informed Trees (AIT*): Fast Asymptotically Optimal Path Planning through Adaptive Heuristics](https://mlsanity.com/paper/2002.06599/) (2020)
- [Geometric Fabrics: Generalizing Classical Mechanics to Capture the Physics of Behavior](https://mlsanity.com/paper/2109.10443/) (2021)
- [Neural MP: A Generalist Neural Motion Planner](https://mlsanity.com/paper/2409.05864/) (2024)
- [A New Approach to Time-Optimal Path Parameterization based on Reachability Analysis](https://mlsanity.com/paper/1707.07239/) (2017)
- [The Convex Feasible Set Algorithm for Real Time Optimization in Motion Planning](https://mlsanity.com/paper/1709.00627/) (2017)
- [Deep Reactive Policy: Learning Reactive Manipulator Motion Planning for Dynamic Environments](https://mlsanity.com/paper/2509.06953/) (2025)
- [Learning on the Fly: Rapid Policy Adaptation via Differentiable Simulation](https://mlsanity.com/paper/2508.21065/) (2025)
- [Learning Fast, Tool aware Collision Avoidance for Collaborative Robots](https://mlsanity.com/paper/2508.20457/) (2025)

## Abstract (arXiv)

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.