Driving · 2026-11-03
X-World 1.6: Building Controllable Virtual Cities with World Models
XPeng, :, Boyang Wang, Boyi Sun, Chaoda Zheng, Dongbin Zhang, Guangfeng Jiang, Hanlong Chen, Hongbin Lin, Jianlei Zheng, Jinhao Deng, Kai Yang, Kangjie Chen, Linkun Xu, Minqi Wang, Puyi Wang, Rongliang Cheng, Ruixin Liu, Shijia Chen, Shuang Deng, Sijia Chen, Sin Wai Choo, Tianyuan Xie, Xiangyu Li, Xianming Liu, Xingjie Liu, Xinlong Zheng, Xuejie Lv, Xuezhi Yang, Yi Guo, Yixiao Zeng, Yu Zhang, Yuecheng Ma, Yuhao Wang, Yuqi Zhang, Yuxin Peng
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
X-World 1.6 advances X-World 1.0 toward a more reliable, controllable driving simulator. The authors add finer-grained traffic semantics control (traffic lights and road markings) and improved spatial consistency across cameras. They introduce absolute static scene control (point clouds, 3D Gaussian Splatting, or reference images), an enhanced long-horizon post-training pipeline (Connected Self Forcing), and a dedicated evaluation/iteration framework.
Why it matters
Generative world models for closed-loop evaluation and policy learning require precise scenario control and consistent observations during long interactions. X-World 1.6 expands controllable multi-camera scenario construction (Virtual City) and improves long-horizon causal generation quality via post-training and reward refinement, plus evaluation infrastructure to guide iteration.
Method
- Improved base-model controllability: added traffic-light conditions (spatial localization with per-source-frame state tokens) and discrete road-surface markings, plus absolute static scene control with 3D point clouds, 3DGS references, or reference images.
- Geometry-aware multi-view architecture: geometry-aware positional encoding (GeoPE) and joint self-attention across time and views in a DiT backbone; separate conditioning branches fused after cross-view attention.
- Long-horizon post-training: retain two-stage training, use Connected Self Forcing (CSF) to restore feedback from later predictions through history on which they depend, extend teacher supervision/student rollouts, and reward-guided refinement for visual quality and ego-motion adherence.
Abstract (from arXiv)
Deploying generative world models for end-to-end closed-loop evaluation and policy learning requires precise scenario control, consistent observations, and sustained generation quality during extended interactions. This report presents \emph{X-World~1.6}, advancing X-World~1.0 toward a more reliable and versatile driving simulator. The upgraded base model offers finer-grained control over traffic semantics, including traffic lights and road markings, together with improved spatial consistency across cameras. Virtual City capabilities broaden scenario construction through a family of models conditioned on diverse 2D and 3D references, including support for generating multi-camera future videos from a single image. An enhanced post-training framework enables efficient causal, few-step generation with stronger temporal coherence, better quality retention over longer rollouts, and improved alignment with human preferences. We also introduce a dedicated evaluation suite to support systematic assessment and guide model iteration. Together, these upgrades expand the scenarios X-World can simulate and improve the efficiency and reliability of its generated observations, supporting broader use in autonomous-driving development.
Related papers
- DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving
- GAIA-1: A Generative World Model for Autonomous Driving
- DriveWorld3D Evidence Constrained 3D World Modeling and Generative Simulation for Real-World Autonomous Driving
- Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability
- WAM-Sim0.1: World Anchored Memory Meets Online RL for Consistent and Realistic Driving Simulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-03 · Indexed by Pub-AI: 2026-11-03 UTC
X-World 1.6: Building Controllable Virtual Cities with World Models
First announced by arXiv: 2026-11-03 · Original record ↗
- arxiv
- 2611.00596
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-03 · First detected: 2026-11-03 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.00596v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
Deploying generative world models for end-to-end closed-loop evaluation and policy learning requires precise scenario control, consistent observations, and sustained generation quality during extended interactions. This report presents \emph{X-World~1.6}, advancing X-World~1.0 toward a more reliable and versatile driving simulator. The upgraded base model offers finer-grained control over traffic semantics, including traffic lights and road markings, together with improved spatial consistency across cameras. Virtual City capabilities broaden scenario construction through a family of models conditioned on diverse 2D and 3D references, including support for generating multi-camera future videos from a single image. An enhanced post-training framework enables efficient causal, few-step generation with stronger temporal coherence, better quality retention over longer rollouts, and improved alignment with human preferences. We also introduce a dedicated evaluation suite to support systematic assessment and guide model iteration. Together, these upgrades expand the scenarios X-World can simulate and improve the efficiency and reliability of its generated observations, supporting broader use in autonomous-driving development.
