← Back ICRA 2026

Adaptive Linear Path Model-Based Diffusion

Yutaka Shimizu, Masayoshi Tomizuka

PDF

AI summary

Key figure (auto-extracted from paper)

Decoupling noise scale from diffusion steps via a linear probability path enables reinforcement learning to dynamically adapt scheduling parameters, significantly improving the robustness and efficiency of diffusion-based robotic control.

Diffusion models model-based control trajectory optimization reinforcement learning adaptive scheduling robotic planning

Problem

Model-based diffusion planners for robotics suffer from tightly coupled noise scheduling parameters that require tedious, task-specific tuning and cannot adapt to varying environmental complexity.

Approach

The authors replace variance-preserving schedules with a linear probability path to decouple noise magnitude from step count, then train a reinforcement learning agent to dynamically adjust these parameters based on real-time task conditions.

Key results

LP-MBD decouples maximum noise scale from diffusion steps, simplifying hyperparameter tuning
ALP-MBD uses reinforcement learning to dynamically adjust scheduling parameters in real-time
LP-MBD converges faster and more reliably than variance-preserving baselines on numerical benchmarks
ALP-MBD improves robustness, adaptability, and real-time efficiency across simulations and mobile robot tracking

Why it matters

It enables more reliable and efficient deployment of diffusion-based planners in robotics by eliminating tedious parameter tuning and allowing real-time adaptation to complex environments.

Abstract

The interest in combining model-based control approaches with diffusion models has been growing. Although we have seen many impressive robotic control results in difficult tasks, the performance of diffusion models is highly sensitive to the choice of scheduling parameters, making parameter tuning one of the most critical challenges. We introduce Linear Path Model-Based Diffusion (LP-MBD), which replaces the variance-preserving schedule with a flow-matching–inspired linear probability path. This yields a geometrically interpretable and decoupled parameterization that reduces tuning complexity and provides a stable foundation for adaptation. Building on this, we propose Adaptive LP-MBD (ALP-MBD), which leverages reinforcement learning to adjust diffusion steps and noise levels according to task complexity and environmental conditions. Across numerical studies, Brax benchmarks, and mobile-robot trajectory tracking, LP-MBD simplifies schedul- ing while maintaining strong performance, and ALP-MBD further improves robustness, adaptability, and real-time effi- ciency. Our code is available through anonymous repository https://anonymous.4open.science/r/adaptive_l inear_path_model_based_diffusion-C58C

Index terms

Machine Learning for Robot Control Optimization and Optimal Control Motion and Path Planning