Adaptive Linear Path Model-Based Diffusion

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 scheduling while maintaining strong performance, and ALP-MBD further improves robustness, adaptability, and real-time efficiency.

💡 Research Summary

This paper addresses a practical bottleneck in Model‑Based Diffusion (MBD) for robotic trajectory optimization: the difficulty of tuning the noise schedule. Traditional MBD relies on a variance‑preserving (VP) schedule parameterized by three tightly coupled values (β₀, β₁, T). Changing any of them simultaneously alters the effective noise level and the discretization, making systematic tuning labor‑intensive and often sub‑optimal across varying task conditions.

The authors propose Linear Path Model‑Based Diffusion (LP‑MBD), which replaces the VP schedule with a linear probability path inspired by Flow Matching. Specifically, they define a continuous interpolation Yₜ = (1‑t)Y⁰ + tε for t∈