# KVPO: ODE-Native GRPO for Autoregressive Video Alignment via KV Semantic Exploration

arXiv: https://arxiv.org/abs/2605.14278 · HTML: https://mlsanity.com/paper/2605.14278/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 22 of its 33 references that are in the map.
- First posted on arXiv 14 May 2026.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)
- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)

## Builds on (most cited first, 22 of 22)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Flow-GRPO: Training Flow Matching Models via Online RL](https://mlsanity.com/paper/2505.05470/) (2025)
- [Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion](https://mlsanity.com/paper/2506.08009/) (2025)
- [DanceGRPO: Unleashing GRPO on Visual Generation](https://mlsanity.com/paper/2505.07818/) (2025)
- [VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness](https://mlsanity.com/paper/2503.21755/) (2025)
- [Improving Video Generation with Human Feedback](https://mlsanity.com/paper/2501.13918/) (2025)
- [LongLive: Real-time Interactive Long Video Generation](https://mlsanity.com/paper/2509.22622/) (2025)
- [HPSv3: Towards Wide-Spectrum Human Preference Score](https://mlsanity.com/paper/2508.03789/) (2025)
- [Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation](https://mlsanity.com/paper/2512.04678/) (2025)
- [MemFlow: Flowing Adaptive Memory for Consistent and Efficient Long Video Narratives](https://mlsanity.com/paper/2512.14699/) (2025)
- [UniVid: The Open-Source Unified Video Model](https://mlsanity.com/paper/2509.24200/) (2025)
- [Neighbor GRPO: Contrastive ODE Policy Optimization Aligns Flow Models](https://mlsanity.com/paper/2511.16955/) (2025)
- [Astrolabe: Steering Forward-Process Reinforcement Learning for Distilled Autoregressive Video Models](https://mlsanity.com/paper/2603.17051/) (2026)
- [Manifold-Aware Exploration for Reinforcement Learning in Video Generation](https://mlsanity.com/paper/2603.21872/) (2026)
- [Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models](https://mlsanity.com/paper/2605.09681/) (2026)
- [RoboStereo: Dual-Tower 4D Embodied World Models for Unified Policy Optimization](https://mlsanity.com/paper/2603.12639/) (2026)
- [MIND-V: Hierarchical World Model for Long-Horizon Robotic Manipulation with RL-based Physical Alignment](https://mlsanity.com/paper/2512.06628/) (2025)
- [Beyond VLM-Based Rewards: Diffusion-Native Latent Reward Modeling](https://mlsanity.com/paper/2602.11146/) (2026)
- [AR-CoPO: Align Autoregressive Video Generation with Contrastive Policy Optimization](https://mlsanity.com/paper/2603.17461/) (2026)
- [Identity-Consistent Video Generation under Large Facial-Angle Variations](https://mlsanity.com/paper/2603.21299/) (2026)

## Built on by (most cited first, 11 of 11)

- [SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models](https://mlsanity.com/paper/2609.02886/) (2026)
- [When Should the Teacher Move? Temporal Coupling and Stability in Self On-Policy Distillation](https://mlsanity.com/paper/2606.03532/) (2026)
- [Embedding-perturbed Exploration Preference Optimization for Flow Models](https://mlsanity.com/paper/2605.15803/) (2026)
- [TempAct: Advancing Temporal Plausibility in Autoregressive Video Generation via Planner-Executor RL](https://mlsanity.com/paper/2606.28016/) (2026)
- [Scaling Reinforcement Learning for Diffusion Models via Velocity Matching](https://mlsanity.com/paper/2608.23664/) (2026)
- [Learning Visual Spatial Planning from Symbolic State via Modality-Gap-Aware Self-Distillation](https://mlsanity.com/paper/2606.06076/) (2026)
- [TexTailor: Texture-Preserving Video Virtual Try-On via Adaptive Garment Conditioning](https://mlsanity.com/paper/2609.39335/) (2026)
- [RADC: Risk-Aware Dual Caching for Vision-Language Test-Time Adaptation](https://mlsanity.com/paper/2610.06932/) (2026)
- [Memory Forcing: Attendable Mid-Horizon History for Streaming Video Generation](https://mlsanity.com/paper/2610.11756/) (2026)
- [ContourVLA: A Closed-Loop Perception-Action Contour Policy for Generalized Referring Expression Segmentation](https://mlsanity.com/paper/2610.12107/) (2026)
- [Self-Verifier Capture in Unified Visual Generators: Diagnosing and Repairing Closed-Loop Self-Reward](https://mlsanity.com/paper/2610.16500/) (2026)

## Abstract (arXiv)

Aligning streaming autoregressive (AR) video generators with human preferences is challenging. Existing reinforcement learning methods predominantly rely on noise-based exploration and SDE-based surrogate policies that are mismatched to the deterministic ODE dynamics of distilled AR models, and tend to perturb low-level appearance rather than the high-level semantic storyline progression critical for long-horizon coherence. To address these limitations, we present KVPO, an ODE-native online Group Relative Policy Optimization (GRPO) framework for aligning streaming video generators. For diversity exploration, KVPO introduces a causal-semantic exploration paradigm that relocates the source of variation from stochastic noise to the historical KV cache. By stochastically routing historical KV entries, it constructs semantically diverse generation branches that remain strictly on the data manifold. For policy modeling, KVPO introduces a velocity-field surrogate policy based on Trajectory Velocity Energy (TVE), which quantifies branch likelihood in flow-matching velocity space and yields a reward-weighted contrastive objective fully consistent with the native ODE formulation. Experiments on multiple distilled AR video generators demonstrate consistent gains in visual quality, motion quality, and text-video alignment across both single-prompt short-video and multi-prompt long-video settings.