AI News AI资讯 2d ago Updated 2d ago 更新于 2天前 45

Niantic Spatial Looks to Close the Robotics Sim-to-Real Gap with USDZ Export in Scaniverse Niantic Spatial 希望通过 Scaniverse 中的 USDZ 导出功能缩小机器人仿真到现实的差距

Niantic Spatial has integrated USDZ export into Scaniverse, enabling direct conversion of real-world 360-degree video captures into simulation-ready environments for Nvidia Isaac Sim. The workflow combines Gaussian splats for high-fidelity visual rendering with automatically generated, aligned meshes for physical collision geometry, addressing the sim-to-real gap. This solution lowers barriers to entry by utilizing affordable $500 360-degree cameras and rapid five-minute capture times, replacing Niantic Spatial在Scaniverse中新增USDZ导出功能,支持将360度视频捕获转化为Nvidia Isaac Sim可用的仿真环境。 该功能结合高斯泼溅(Gaussian Splatting)与自动生成的对齐网格,保留光照纹理并生成碰撞几何体,以缩小Sim-to-Real差距。 使用低至500美元的360度相机即可在几分钟内完成场景扫描,替代昂贵的RGB-LiDAR设备,降低机器人训练成本。 从泼溅数据直接派生网格确保了视觉层与物理层的精确对齐,避免了单独配准步骤,提升了数字孪生的保真度。 生成的数字孪生环境既可用于机器人初始策略训练,也可用于部署后的策略优化与微调。

65
Hot 热度
60
Quality 质量
65
Impact 影响力

Analysis 深度分析

TL;DR

  • Niantic Spatial has integrated USDZ export into Scaniverse, enabling direct conversion of real-world 360-degree video captures into simulation-ready environments for Nvidia Isaac Sim.
  • The workflow combines Gaussian splats for high-fidelity visual rendering with automatically generated, aligned meshes for physical collision geometry, addressing the sim-to-real gap.
  • This solution lowers barriers to entry by utilizing affordable $500 360-degree cameras and rapid five-minute capture times, replacing expensive RGB-LiDAR setups.
  • The unified export ensures visual and physical layers remain perfectly aligned without separate registration steps, supporting both initial training and post-deployment policy refinement.

Why It Matters

This development significantly democratizes access to high-quality digital twin creation for robotics, allowing developers to bypass costly professional scanning equipment while maintaining simulation accuracy. By streamlining the pipeline from real-world capture to Nvidia Isaac Sim, it accelerates the iteration cycle for robot training, making it easier to deploy policies in complex, unstructured environments.

Technical Details

  • Dual-Layer Export: The system generates a single USDZ file containing both a Gaussian splat (for visual fidelity, lighting, and texture) and an aligned mesh (for collision geometry, surface topology, and elevation).
  • Alignment Mechanism: The mesh is derived directly from the Gaussian splat, ensuring automatic alignment between visual and physical layers and eliminating the need for separate registration processes.
  • Hardware Accessibility: The workflow supports capture via consumer-grade 360-degree cameras costing as little as $500, with full scene capture achievable in approximately five minutes.
  • Integration Pipeline: The process involves capturing data with a 360 camera, uploading to Scaniverse web for processing, exporting as USDZ, and importing into Nvidia Isaac Sim or Isaac Lab for policy training.

Industry Insight

  • Cost Reduction in Robotics Dev: Companies can drastically reduce capital expenditure on scanning hardware by leveraging existing consumer 360 cameras, making digital twin adoption viable for smaller teams.
  • Accelerated Sim-to-Real Transfer: The precise alignment of visual and physical data reduces the domain gap, leading to faster convergence during training and more robust performance in real-world deployments.
  • Standardization of Digital Twins: The move toward USDZ as a standard exchange format between capture tools and simulation engines like Isaac Sim promotes interoperability and simplifies workflow integration across the robotics ecosystem.

TL;DR

  • Niantic Spatial在Scaniverse中新增USDZ导出功能,支持将360度视频捕获转化为Nvidia Isaac Sim可用的仿真环境。
  • 该功能结合高斯泼溅(Gaussian Splatting)与自动生成的对齐网格,保留光照纹理并生成碰撞几何体,以缩小Sim-to-Real差距。
  • 使用低至500美元的360度相机即可在几分钟内完成场景扫描,替代昂贵的RGB-LiDAR设备,降低机器人训练成本。
  • 从泼溅数据直接派生网格确保了视觉层与物理层的精确对齐,避免了单独配准步骤,提升了数字孪生的保真度。
  • 生成的数字孪生环境既可用于机器人初始策略训练,也可用于部署后的策略优化与微调。

为什么值得看

这项更新为机器人开发者提供了一条低成本、高效率的路径,将现实世界环境快速转化为高保真仿真环境,显著降低了Sim-to-Real迁移的技术门槛。它通过整合视觉感知与物理碰撞数据,解决了传统仿真中视觉与几何分离的痛点,对于加速具身智能落地具有重要参考价值。

技术解析

  • 核心工作流:使用360度相机捕获环境 -> 上传至Scaniverse Web生成高斯泼溅和对齐网格 -> 导出为单一USDZ文件 -> 导入Nvidia Isaac Sim/Lab进行训练。
  • 技术融合:利用高斯泼溅保留真实场景的光照、纹理和视觉杂乱信息,满足基于摄像头的机器人感知需求;同时生成网格作为物理层,提供碰撞几何、表面拓扑和高度变化数据。
  • 对齐优势:网格直接从泼溅数据派生,确保视觉与物理层在单次捕获中即已对齐,消除了传统流程中额外的配准步骤,提高了几何表面的平滑度和准确性。
  • 成本效益:相比昂贵的RGB-LiDAR套装,仅需约500美元的360度相机即可完成大型室内或室外场景的快速数字化(约5分钟),大幅降低了数据采集硬件成本。

行业启示

  • 降低具身智能部署门槛:低成本、快速的场景数字化方案使得中小型企业也能构建高质量的机器人训练环境,加速了仿真驱动的开发范式普及。
  • Sim-to-Real闭环优化:强调视觉保真度与物理准确性的统一,表明未来的机器人训练将更依赖于高保真的数字孪生,以减少现实部署中的适应性问题。
  • 工具链标准化趋势:USDZ作为通用交换格式在机器人仿真中的应用深化,促进了不同采集工具(如Scaniverse)与仿真平台(如Isaac Sim)之间的无缝集成。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Robotics 机器人 Product Launch 产品发布