QuantWAMs: Calibrating at the Right Granularity for World Action Models

2026/07/30 by Jiacheng Zhou, Jinfan Lv, Ruixuan Li +4
Computer Science · #cs.AI #cs.LG

paper · pdf

13 pages, 6 figures

arxiv created 2026/07/30 · arxiv updated 2026/07/31

Abstract

World Action Models (WAMs) jointly predict future observations and actions, but their iterative denoising and closed-loop execution make efficient deployment costly. Existing post-training quantization (PTQ) methods are poorly suited to WAMs because they rely on open-loop objectives, homogeneous model assumptions, and calibration distributions that do not reflect deployment. We present QuantWAMs, a PTQ framework that aligns quantization decisions with the calibration context defined by model structure, rollout distribution, and task objective. QuantWAMs introduces three strategies: shared-basis outlier calibration, which pools activation evidence only across coordinate-compatible modules; co-training-objective saliency, which computes empirical-Fisher scores from the joint video--action gradient and assigns weight precision at a calibration-stable layer granularity; and fixed-intervention rollout auditing, which revises denoising-step protection schedules using reachable closed-loop states without changing the precision budget. We evaluate QuantWAMs on Fast-WAM and LingBot-VA across RoboTwin 2.0, LIBERO, and real-robot manipulation with an AgiBot G2. Under a W4A4-dominant setting, the reported simulation means differ from FP16 by 0.2--0.7 percentage points. Real-robot trials further establish deployment feasibility on three manipulation tasks. For the targeted video and action blocks, QuantWAMs reduces peak weight-and-activation memory to about 29% of FP16 and provides 1.4--1.6× block-level speedups.

Citations

• Motus: A Unified Latent Action World Model
• Video Generators are Robot Policies
• WorldVLA: Towards Autoregressive Action World Model
• RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
• Saliency-Aware Quantized Imitation Learning for Efficient Robotic Control
• Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers
• ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation
• Atom: Low-bit Quantization for Efficient and Accurate LLM Serving
• LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
• Temporal Dynamic Quantization for Diffusion Models
• AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
• PTQD: Accurate Post-Training Quantization for Diffusion Models
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
• Diffusion policy: Visuomotor policy learning via action diffusion
• Q-Diffusion: Quantizing Diffusion Models
• Scalable Diffusion Models with Transformers
• Post-training Quantization on Diffusion Models
• SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
• GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
• Optimizing Neural Networks with Kronecker-factored Approximate Curvature
• A Reduction of Imitation Learning and Structured Prediction to No-Regret\n Online Learning

Related

• Test-Time Scaling for World Action Models via Zero-Shot Geometric Evaluation
• GigaWorld-Policy-0.5: A Faster and Stronger WAM Empowered by AutoResearch
• Do World Action Models Generalize Better than VLAs? A Robustness Study
• VQ-VLA: Improving Vision-Language-Action Models via Scaling Vector-Quantized Action Tokenizers
• Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach