Searcharxiv⌕ Search

arXiv subjects

Yuhu Cheng

Publications and source records attributed to Yuhu Cheng.

3 recordsLinked to original sources

Marginal-Preserving Coupled Diffusion Composition for Long-Horizon Planning

Compositional diffusion planners extend short-horizon trajectory models to long-horizon goal-conditioned planning by generating and stitching multiple local trajectory chunks. However, in multimodal environments, local plausibility is not sufficient for global consistency: independently initialized chunks may denoise toward incompatible overlap modes, producing plans that are individually feasible but incoherent after stitching. Existing remedies typically add search, guidance, pruning, or graph-based structure at inference time, increasing computational cost or relying on task-specific machinery. We instead investigate graph-structured dependence among chunk-wise initial diffusion noises as a lightweight coordination mechanism for compositional trajectory planning. We propose \textbf{M}arginal-Preserving \textbf{C}oupled \textbf{D}iffusion \textbf{C}omposition (MCDC), a training-free inference method for pretrained compositional diffusion planners. Instead of sampling chunk noises independently, MCDC constructs coupled Gaussian initializations whose per-chunk marginals remain exactly standard Gaussian, as required by the pretrained denoiser, while neighboring chunks receive positively correlated noise components. This preserves the standard-Gaussian initialization marginal of each chunk while biasing adjacent chunks toward compatible overlap-mode choices. We prove exact marginal preservation and show, under a simple multimodal overlap model, that positive coupling increases the probability of compatible mode selection. On OGBench long-horizon stitch tasks, MCDC consistently improves over independent compositional diffusion sampling while retaining nearly the same inference time, and achieves a favorable success--cost trade-off compared with search-based composition methods.

cs.AI↗

Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning

Offline goal-conditioned reinforcement learning (GCRL) learns goal-directed policies from reward-free data, but in long-horizon tasks, goal-conditioned value functions often provide unstable guidance due to sparse rewards and discounting. Hierarchical methods partially mitigate this issue via subgoal decomposition; however, high-level decision-making still relies on noise-sensitive value estimates, leading to unstable behavior in complex environments. We address this limitation by proposing \textbf{D}iffusion \textbf{S}ubgoal \textbf{P}lanning (\textbf{DSP}), a diffusion-based framework for high-level subgoal generation. DSP casts high-level planning as guided generative inference over goal-conditioned subgoals and learns both conditional and unconditional flows, enabling classifier-free guidance to introduce a goal-directed bias at inference time. By removing explicit value-based guidance from high-level planning, DSP generates reachable and goal-directed subgoals through a generative model while retaining hierarchical execution. Experiments on offline GCRL benchmarks demonstrate that DSP outperforms prior methods on a range of navigation and manipulation tasks, with particularly strong performance in maze environments that require multi-step subgoal planning.

cs.AI↗

Dispersion-Aware Modeling Framework for Parallel Optical Computing

Optical computing represents a groundbreaking technology that leverages the unique properties of photons, with innate parallelism standing as its most compelling advantage. Parallel optical computing like cascaded Mach-Zehnder interferometers (MZIs) based offers powerful computational capabilities but also introduces new challenges, particularly concerning dispersion due to the introduction of new frequencies. In this work, we extend existing theories of cascaded MZI systems to develop a generalized model tailored for wavelength-multiplexed parallel optical computing. Our comprehensive model incorporates component dispersion characteristics into a wavelength-dependent transfer matrix framework and is experimentally validated. We propose a computationally efficient compensation strategy that reduces global dispersion error within a 40 nm range from 0.22 to 0.039 using edge-spectrum calibration. This work establishes a fundamental framework for dispersion-aware model and error correction in MZI-based parallel optical computing chips, advancing the reliability of multi-wavelength photonic processors.

physics.optics↗