WAM-Cache: Staleness-Bounded KV Reuse for Efficient World Action Models
cs.RO, cs.CV, cs.LG
Submitted: 2026-10-08
Updated: 2026-10-08
Comments: 19 pages, 5 figures, 7 tables. Project page: https://dingkai0302.github.io/wam-cache/
Project page: https://dingkai0302.github.io/wam-cache/ABSTRACT
License: http://creativecommons.org/licenses/by/4.0/
The gist: World Action Models (WAMs) enable generalist robot manipulation by conditioning an action expert on representations from a pretrained video Diffusion Transformer (DiT).
Terminology
Abstract
World Action Models (WAMs) enable generalist robot manipulation by conditioning an action expert on representations from a pretrained video Diffusion Transformer (DiT). In closed-loop control, the video DiT runs at every chunk to encode the current observation into layerwise key-value (KV) pairs that the action expert queries. This prefill dominates the per-chunk computational cost, yet existing training-free accelerations leave it fully dense. We present WAM-Cache, a training-free framework that retains layerwise key-value representations across chunks and recomputes only a sparse refresh set of tokens. Crucially, we find that the intuitive heuristic of refreshing visually drifted tokens plateaus far below the dense baseline, even with an oracle predicting ground-truth KV drift. Downstream action accuracy is instead governed by where the action expert attends, not by what moved. WAM-Cache therefore selects the refresh set by uniting the action expert's cross-attention with visual latent surprise, complemented by a strict age bound that suppresses compounding error. On Fast-WAM, WAM-Cache cuts video DiT prefill FLOPs by 32-42% across RoboTwin 2.0, LIBERO, and real-world experiments, while staying within 0.7-1.8 percentage points of the dense policy in simulation and 2.5 points on a real robot.
Sources
- Flash-WAM: Modality-Aware Distillation for World Action Models
- Gen2Act: Human Video Generation in Novel Scenarios enables Generalizable Robot Manipulation
- Motus: A Unified Latent Action World Model
- $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control
- Token Merging: Your ViT But Faster
- RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control
- WorldVLA: Towards Autoregressive Action World Model
- GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation
- LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies
- An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models
- RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
- High-Fidelity One-Step Generative Visuomotor Policy via Recursive Correction, Frequency Consistency, and Contrastive Flow Matching
- From Flow to One Step: Real-Time Multi-Modal Trajectory Policies via Implicit Maximum Likelihood Estimation-based Distribution Distillation
- Learning Universal Policies via Text-Guided Video Generation
- Eventful Transformers: Leveraging Temporal Redundancy in Vision Transformers
- WorldCache: Accelerating World Models for Free via Heterogeneous Token Caching
- Reflex: Real-Time VLA Control through Streaming Inference
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations
- Long-WAM: Scaling the Context of World-Action Models
- DreamGen: Unlocking Generalization in Robot Learning through Video World Models
Related papers
- FMT x: An Efficient and Asymptotically Optimal Extension of the Fast Marching Tree for Dynamic Replanning
- MPCFormer: A physics-informed data-driven approach for explainable socially-aware autonomous driving
- RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies
- ARC: A Reasoning Recipe for Robot Foundation Models
- HRDexDB: A 4D Dexterous Grasping Dataset Across Human and Multiple Robot Embodiments
- APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies