Wenjian Zhang
Publications
Rectified Schrödinger Bridge Matching for Few-Step Visual Navigation
Visual navigation is a core challenge in Embodied AI, requiring autonomous agents to translate high-dimensional sensory observations into continuous, long-horizon action trajectories. While generative policies based on diffusion models and Schrödinger Bridges (SB) effectively capture multimodal action distributions, they require dozens of integration steps due to high-variance stochastic transport, posing a critical barrier for real-time robotic control. We propose Rectified Schrödinger Bridge Matching (RSBM), a framework that exploits a shared velocity-field structure between standard Schrödinger Bridges ($\varepsilon=1$, maximum-entropy transport) and deterministic Optimal Transport ($\varepsilon\to 0$, as in Conditional Flow Matching), controlled by a single entropic regularization parameter $\varepsilon$. We prove two key results: (1) the conditional velocity field's functional form is invariant across the entire $\varepsilon$-spectrum (Velocity Structure Invariance), enabling a single network to serve all regularization strengths; and (2) reducing $\varepsilon$ linearly decreases the conditional velocity variance, enabling more stable coarse-step ODE integration. Anchored to a learned conditional prior that shortens transport distance, RSBM operates at an intermediate $\varepsilon$ that balances multimodal coverage and path straightness. Empirically, while standard bridges require $\geq 10$ steps to converge, RSBM achieves over 94% cosine similarity and 92% success rate in merely 3 integration steps -- without distillation or multi-stage training -- substantially narrowing the gap between high-fidelity generative policies and the low-latency demands of Embodied AI.
SnapFlow: One-Step Action Generation for Flow-Matching VLAs via Progressive Self-Distillation
Vision-Language-Action (VLA) models based on flow matching -- such as pi0, pi0.5, and SmolVLA -- achieve state-of-the-art generalist robotic manipulation, yet their iterative denoising, typically 10 ODE steps, introduces substantial latency: on a modern GPU, denoising alone accounts for 80% of end-to-end inference time. Naively reducing the step count is unreliable, degrading success on most tasks due to the velocity field being uncalibrated for single-step jumps. We present SnapFlow, a plug-and-play self-distillation method that compresses multi-step denoising into a single forward pass (1-NFE) for flow-matching VLAs. SnapFlow mixes standard flow-matching samples with consistency samples whose targets are two-step Euler shortcut velocities computed from the model's own marginal velocity predictions, avoiding the trajectory drift caused by conditional velocities, as we analyze theoretically. A zero-initialized target-time embedding lets the network switch between local velocity estimation and global one-step generation within a single architecture. SnapFlow requires no external teacher, no architecture changes, and trains in ~12h on a single GPU. We validate on two VLA architectures spanning a 6x parameter range, with identical hyperparameters: on pi0.5 (3B) across four LIBERO suites (40 tasks, 400 episodes), SnapFlow achieves 98.75% average success -- matching the 10-step teacher at 97.75% and slightly exceeding it -- with 9.6x denoising speedup and end-to-end latency reduced from 274ms to 83ms; on SmolVLA (500M), it reduces MSE by 8.3% with 3.56x end-to-end acceleration. An action-step sweep on long-horizon tasks reveals that SnapFlow maintains its advantage across execution horizons, achieving 93% at n_act=5 where the baseline reaches only 90%. SnapFlow is orthogonal to layer-distillation and token-pruning approaches, enabling compositional speedups.
Experience is the Best Teacher: Motivating Effective Exploration in Reinforcement Learning for LLMs
Reinforcement Learning (RL) with rubric-based rewards has recently shown remarkable progress in enhancing general reasoning capabilities of Large Language Models (LLMs), yet still suffers from ineffective exploration confined to curent policy distribution. In fact, RL optimization can be viewed as steering the policy toward an ideal distribution that maximizes the rewards, while effective exploration should align efforts with desired target. Leveraging this insight, we propose HeRL, a Hindsight experience guided Reinforcement Learning framework to bootstrap effective exploration by explicitly telling LLMs the desired behaviors specified in rewards. Concretely, HeRL treats failed trajectories along with their unmet rubrics as hindsight experience, which serves as in-context guidance for the policy to explore desired responses beyond its current distribution. Additionally, we introduce a bonus reward to incentivize responses with greater potential for improvement under such guidance. HeRL facilitates effective learning from desired high quality samples without repeated trial-and-error from scratch, yielding a more accurate estimation of the expected gradient theoretically. Extensive experiments across various benchmarks demonstrate that HeRL achieves superior performance gains over baselines, and can further benefit from experience guided self-improvement at test time. Our code is available at https://github.com/sikelifei/HeRL.