來源36氪•較早收集於 5m
五一視界SimOne 4.0先行版發布，深度融合英偉達產品
💡英偉達融合仿真平台，具身AI關鍵工具–機器人開發者擴展物理訓練必讀 (38字)
⚡ 30 秒速覽
有什麼變化
以世界模型和VLA重構，助力AI進入物理世界
為什麼重要
加速機器人和自動駕駛系統的安全AI部署，提供可擴展仿真。強化英偉達生態系統對具身AI開發者的支持。定位中國仿真平台為全球競爭者。
下一步行動
下載SimOne 4.0先行版，測試Omniverse NuRec整合用於機器人訓練管道。
誰應關注：Developers & AI Engineers
關鍵要點
- •以世界模型和VLA重構，助力AI進入物理世界
- •完成與英偉達Omniverse NuRec深度融合
- •完整流程：真實數據採集、神經場景重建、閉環仿真
- •9年迭代，從SimOne 1.0感知仿真到生成技術
🧠 深度解析
本篇為 AI 生成分析，非原文內容。
🔑 增強重點摘要
- •SimOne 4.0 shifts from traditional rule-based simulation to a generative AI paradigm, utilizing VLA (Vision-Language-Action) models to enable autonomous agents to interpret and interact with synthetic environments in real-time.
- •The integration with Nvidia Omniverse NuRec (Neural Reconstruction) allows for the automated conversion of raw sensor data from real-world driving scenarios into high-fidelity, photorealistic 3D digital twins, significantly reducing manual modeling labor.
- •The platform now supports 'Scenario-as-Code' capabilities, allowing developers to generate complex, edge-case traffic scenarios via natural language prompts, accelerating the validation cycle for Level 4 autonomous driving systems.
📊 競品分析▸ Show
| Feature | SimOne 4.0 | Applied Intuition | dSPACE |
|---|---|---|---|
| Core Tech | World Models/VLA | Simulation/Data Engine | Hardware-in-the-loop |
| Nvidia Integration | Deep (Omniverse/NuRec) | Moderate | Limited |
| Primary Focus | Generative Reconstruction | End-to-end AV Dev | Validation/Testing |
| Pricing | Enterprise/Custom | Enterprise/SaaS | Enterprise/Hardware |
🛠️ 技術深入
- •Architecture: Built on a transformer-based world model backbone capable of predicting future states of dynamic objects (vehicles, pedestrians) based on historical sensor sequences.
- •VLA Integration: Implements a VLA policy head that maps visual inputs directly to control commands, allowing the simulation to test agent decision-making in non-deterministic environments.
- •Data Pipeline: Utilizes a neural radiance field (NeRF) or Gaussian Splatting-based reconstruction engine within the NuRec framework to achieve sub-centimeter accuracy in static environment reconstruction.
- •Closed-loop Simulation: Supports high-frequency synchronization between the physics engine and the neural reconstruction layer to ensure consistent collision detection and sensor feedback.
🔮 前景展望基於引用來源的 AI 分析
SimOne 4.0 will reduce AV training data costs by over 40%.
By automating the generation of synthetic edge cases through generative reconstruction, companies can rely less on expensive, manual real-world data collection.
The platform will become a standard for 'Sim-to-Real' transfer in China's AV industry.
Deep integration with Nvidia's ecosystem provides a standardized, high-performance infrastructure that aligns with the hardware stacks used by major Chinese EV manufacturers.
⏳ 時間線
2017-05
五一視界 (51World) founded, initiating the development of SimOne 1.0.
2019-09
SimOne 2.0 released, focusing on high-precision map integration and perception simulation.
2022-06
SimOne 3.0 launched, introducing cloud-native simulation and large-scale parallel testing capabilities.
2026-04
SimOne 4.0 preview announced, marking the transition to world models and generative reconstruction.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： 36氪 ↗
每週電子報
每週一封，可隨時退訂。