Sample from What You See: Visuomotor Policy Learning via Diffusion Bridge with Observation-Embedded Stochastic Differential Equation

2025-12-08

AI Summary

Existing imitation learning methods treat observations solely as conditional inputs to diffusion denoising networks, initializing sampling from random Gaussian noise—thereby decoupling perception from control and limiting performance. This paper proposes BridgePolicy, the first approach to directly embed visual and state observations into the stochastic differential equation (SDE) dynamics of the diffusion process, enabling generative policy modeling grounded in information priors rather than noise. To address heterogeneity between multimodal observations and action spaces, we design a multimodal fusion module and a semantic aligner. Evaluated on three simulation benchmarks (52 tasks) and five real-world robotic tasks, BridgePolicy significantly outperforms state-of-the-art methods, demonstrating superior control accuracy, robustness to distribution shifts, and cross-scenario generalization capability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Imitation learning with diffusion models has advanced robotic control by capturing multi-modal action distributions. However, existing approaches typically treat observations as high-level conditioning inputs to the denoising network, rather than integrating them into the stochastic dynamics of the diffusion process itself. As a result, sampling must begin from random Gaussian noise, weakening the coupling between perception and control and often yielding suboptimal performance. We introduce BridgePolicy, a generative visuomotor policy that explicitly embeds observations within the stochastic differential equation via a diffusion-bridge formulation. By constructing an observation-informed trajectory, BridgePolicy enables sampling to start from a rich, informative prior rather than random noise, substantially improving precision and reliability in control. A key challenge is that classical diffusion bridges connect distributions with matched dimensionality, whereas robotic observations are heterogeneous and multi-modal and do not naturally align with the action space. To address this, we design a multi-modal fusion module and a semantic aligner that unify visual and state inputs and align observation and action representations, making the bridge applicable to heterogeneous robot data. Extensive experiments across 52 simulation tasks on three benchmarks and five real-world tasks demonstrate that BridgePolicy consistently outperforms state-of-the-art generative policies.

Problem

• Integrates observations into diffusion process dynamics
• Starts sampling from informative prior, not random noise
• Aligns heterogeneous observations with action representations

Innovation

• Embedding observations into diffusion bridge SDE
• Starting sampling from observation-informed prior
• Aligning heterogeneous observations with action space