Selected work

Representative Papers

Mirror Learning

Jul 30, 2026

This work addresses the challenge that existing behavior cloning methods rely on first-person aligned data and struggle to learn effective policies from third-person passive observations. The authors propose a Mirror Learning framework that, for the first time, integrates viewpoint transformation with inverse dynamics modeling. By fine-tuning a video diffusion model to translate third-person observations into first-person perspectives and employing an inverse dynamics model to infer action trajectories, the method generates pseudo-first-person expert demonstrations from purely observational videos. This approach constructs a generative world model capable of training high-performance policies using only mirrored data, substantially reducing reliance on teleoperated demonstrations. When combined with first-person behavior cloning, the framework further enhances downstream policy performance.

0 citationsRead paper

Integration Matters: Rollout-Based Training for Constrained Diffusion Models

Jul 15, 2026

Existing diffusion models often suffer from low constraint satisfaction rates or degraded sample quality when generating samples under complex feasibility constraints, primarily due to distributional mismatches between training and sampling phases. This work proposes a trajectory-aware fine-tuning framework based on online rollouts, which integrates constraint guidance during training and, for the first time, incorporates a numerical integration perspective into the diffusion process. By end-to-end differentiating denoising trajectories under a fixed noise schedule, the method explicitly exposes constraint violations, thereby aligning the training and sampling distributions. Combining constraint-aware guidance, differentiable noise scheduling, and an online rollout mechanism, the approach significantly improves constraint satisfaction across multiple tasks while maintaining generation quality on par with current state-of-the-art methods.

0 citationsRead paper

Filtered Posterior Mean Collections: A Unified Framework for Analytical Models of Diffusion Generalization

May 22, 2026

Existing image diffusion models lack a unified analytical framework for modeling the generalization behavior of neural denoising functions. This work proposes the Filtered Posterior Mean Collections (FPMCs) framework, which, for the first time, unifies various posterior-weighted averaging methods based on training data patches into a single formalism. FPMCs systematically integrates and generalizes these approaches along three design axes: query precision vectors, response weights, and source distributions. To further enhance performance, the framework introduces soft relaxation and source distribution augmentation strategies. Experimental results demonstrate that FPMCs consistently and significantly improve sample quality across three natural image datasets, confirming its effectiveness and broad applicability.

0 citationsRead paper

Improved Constrained Generation by Bridging Pretrained Generative Models

Mar 06, 2026

This work addresses the challenge of enforcing complex nonlinear constraints—such as road-legal regions in robotic control and autonomous driving—within generative models, where existing approaches often fail to simultaneously ensure constraint satisfaction and high-fidelity generation. The authors propose a constrained fine-tuning framework that leverages pre-trained generative models to produce outputs strictly confined within structured feasible regions, without compromising sample realism. By overcoming the limitations of conventional fine-tuning or training-free strategies, the method achieves superior performance across diverse and intricate constraint scenarios, consistently outperforming current baselines in both generation quality and adherence to constraints.

0 citationsRead paper

Constrained Generative Modeling with Manually Bridged Diffusion Models

Feb 27, 2025

Modeling diffusion-based generation under multiple hard constraints (e.g., kinematics, environment) in safety-critical, resource-constrained settings remains challenging. Method: We propose Manual Bridging—a theoretically grounded framework that enables joint embedding and strict satisfaction of heterogeneous hard constraints. We prove it constructs a valid diffusion bridge, ensuring probability flow consistency and constraint completeness; further, we introduce a distribution alignment training strategy to precisely match the data distribution within the constrained space. Contribution/Results: This is the first application of constrained diffusion models to autonomous driving trajectory initialization. Generated trajectories satisfy *all* hard constraints with 100% fidelity; path success rate improves by 23% over SOTA baselines, while feasibility and trajectory diversity are significantly enhanced. Our work establishes a new paradigm for high-reliability generative modeling—rigorous in theory and practical in deployment.

0 citationsRead paper

Recent publications

Latest Papers

Mirror Learning

Jul 30, 2026

This work addresses the challenge that existing behavior cloning methods rely on first-person aligned data and struggle to learn effective policies from third-person passive observations. The authors propose a Mirror Learning framework that, for the first time, integrates viewpoint transformation with inverse dynamics modeling. By fine-tuning a video diffusion model to translate third-person observations into first-person perspectives and employing an inverse dynamics model to infer action trajectories, the method generates pseudo-first-person expert demonstrations from purely observational videos. This approach constructs a generative world model capable of training high-performance policies using only mirrored data, substantially reducing reliance on teleoperated demonstrations. When combined with first-person behavior cloning, the framework further enhances downstream policy performance.

0 citationsRead paper

Integration Matters: Rollout-Based Training for Constrained Diffusion Models

Jul 15, 2026

Existing diffusion models often suffer from low constraint satisfaction rates or degraded sample quality when generating samples under complex feasibility constraints, primarily due to distributional mismatches between training and sampling phases. This work proposes a trajectory-aware fine-tuning framework based on online rollouts, which integrates constraint guidance during training and, for the first time, incorporates a numerical integration perspective into the diffusion process. By end-to-end differentiating denoising trajectories under a fixed noise schedule, the method explicitly exposes constraint violations, thereby aligning the training and sampling distributions. Combining constraint-aware guidance, differentiable noise scheduling, and an online rollout mechanism, the approach significantly improves constraint satisfaction across multiple tasks while maintaining generation quality on par with current state-of-the-art methods.

0 citationsRead paper

Filtered Posterior Mean Collections: A Unified Framework for Analytical Models of Diffusion Generalization

May 22, 2026

Existing image diffusion models lack a unified analytical framework for modeling the generalization behavior of neural denoising functions. This work proposes the Filtered Posterior Mean Collections (FPMCs) framework, which, for the first time, unifies various posterior-weighted averaging methods based on training data patches into a single formalism. FPMCs systematically integrates and generalizes these approaches along three design axes: query precision vectors, response weights, and source distributions. To further enhance performance, the framework introduces soft relaxation and source distribution augmentation strategies. Experimental results demonstrate that FPMCs consistently and significantly improve sample quality across three natural image datasets, confirming its effectiveness and broad applicability.

0 citationsRead paper

Improved Constrained Generation by Bridging Pretrained Generative Models

Mar 06, 2026

This work addresses the challenge of enforcing complex nonlinear constraints—such as road-legal regions in robotic control and autonomous driving—within generative models, where existing approaches often fail to simultaneously ensure constraint satisfaction and high-fidelity generation. The authors propose a constrained fine-tuning framework that leverages pre-trained generative models to produce outputs strictly confined within structured feasible regions, without compromising sample realism. By overcoming the limitations of conventional fine-tuning or training-free strategies, the method achieves superior performance across diverse and intricate constraint scenarios, consistently outperforming current baselines in both generation quality and adherence to constraints.

0 citationsRead paper

Constrained Generative Modeling with Manually Bridged Diffusion Models

Feb 27, 2025

Modeling diffusion-based generation under multiple hard constraints (e.g., kinematics, environment) in safety-critical, resource-constrained settings remains challenging. Method: We propose Manual Bridging—a theoretically grounded framework that enables joint embedding and strict satisfaction of heterogeneous hard constraints. We prove it constructs a valid diffusion bridge, ensuring probability flow consistency and constraint completeness; further, we introduce a distribution alignment training strategy to precisely match the data distribution within the constrained space. Contribution/Results: This is the first application of constrained diffusion models to autonomous driving trajectory initialization. Generated trajectories satisfy *all* hard constraints with 100% fidelity; path success rate improves by 23% over SOTA baselines, while feasibility and trajectory diversity are significantly enhanced. Our work establishes a new paradigm for high-reliability generative modeling—rigorous in theory and practical in deployment.

0 citationsRead paper