mlsanityData to 14 Oct 2026

QuantWAMs: Calibrating at the Right Granularity for World Action Models

arXiv 2607.28405 ↗First posted 30 Jul 2026cs.AI cs.LGcited by 2

On the map: World models → Embodied and robot world models

Abstract (arXiv)

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.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Embodied and robot world models, one of 11 directions in World models. Builds on 24 of its 25 references that are in the map. First posted on arXiv 30 Jul 2026.

Builds on

24 of 25 references.

Closest ancestors · 6

Same direction, newest first

  1. 2026Fast-WAM: Do World Action Models Need Test-time Future Imagination?17 Mar
  2. 2026World Action Models are Zero-shot Policies17 Feb
  3. 2026Causal World Modeling for Robot Control29 Jan
  4. 2025Motus: A Unified Latent Action World Model15 Dec
  5. 2025WorldVLA: Towards Autoregressive Action World Model26 Jun
  6. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusion7 Mar

Other references · 18

  1. 2022Scalable Diffusion Models with Transformerscited by 6,054
  2. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,919
  3. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,410
  4. 2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,386
  5. 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,298
  6. 2015Optimizing Neural Networks with Kronecker-factored Approximate Curvaturecited by 796
  7. 2025RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulationcited by 518
  8. 2023Q-Diffusion: Quantizing Diffusion Modelscited by 273
  9. 2023Atom: Low-bit Quantization for Efficient and Accurate LLM Servingcited by 272
  10. 2022Post-training Quantization on Diffusion Modelscited by 267
  11. 2023PTQD: Accurate Post-Training Quantization for Diffusion Modelscited by 170
  12. 2024ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generationcited by 115
6 more
  1. 2024Q-DiT: Accurate Post-Training Quantization for Diffusion Transformerscited by 94
  2. 2025Video Generators are Robot Policiescited by 92
  3. 2023Temporal Dynamic Quantization for Diffusion Modelscited by 81
  4. 2026QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Modelscited by 30
  5. 2026QVLA: Not All Channels Are Equal in Vision-Language-Action Model's Quantizationcited by 28
  6. 2025Saliency-Aware Quantized Imitation Learning for Efficient Robotic Controlcited by 12

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Q-WAM: 4-Bit Quantization of World Action Models with Action-Subspace Protectioncited by 0
  2. 2026SteerQuant: Steering Quantization Error with Action-Guided Scaling in World-Action Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Predict Before You Deploy: Offline Prediction of Quantization-Induced Task Degradation for World Action Models
  2. 20261Step-WAM: Your World Action Model but Faster using 1-Step Action Generation
  3. 2026An Empirical Study of World Model Quantization
  4. 2026What Makes World Action Models Generalize? An Empirical Study of Test-Time Future Modeling
  5. 2026C3ache: Accelerating World Action Models with Cross Inference Chunk Cache
  6. 2026World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry
  7. 2026DeltaWAM: Distilling Action Tangent Fields into World Action Models
  8. 2026MARS: Systematic Multi-Level Acceleration for World Action Models