预测触觉 token 与由失败、纠正、恢复轨迹构成的 ALTER 离线优势标签已形成可检查的后训练/推理路径：官方固定代码提交含转换、归一化、部署与仿真服务，三份固定权重可下载。但 NeoData、完整预训练、NeoReal 分母、训练算力和独立实机复现仍未公开；因此升级为可受控验证的高潜路线，不视为已完成的人形或跨本体真机复现。
- 权重 · 已核验官方资源
审核码：verified_official; 官方关系：verified_public_nongated_official_huggingface_revision; 最近核验：2026-08-14代码：released separately
权重：public nongated base checkpoint revision
数据：NeoData and pretraining data are not released
许可证：LicenseRef-Gemma-Terms；The model card/repository states checkpoint use inherits Google Gemma Terms and Prohibited Use Policy; repository CC-BY-SA-4.0 does not license weights.
依赖：N0-VTLA fixed source/configuration、PaliGemma/Gemma-compatible dependency path、DINOv2 tactile encoder cache、task-specific post-training data and normalization statistics
硬件/传感器：CUDA-capable GPU、base checkpoint is not itself a deployable task policy
实机证据：A public base checkpoint does not itself establish a physical task replay.
独立复现：No qualifying independent physical checkpoint reproduction was located.
限制/反面证据：Base checkpoint needs task-specific post-training.；No public NeoData or pretraining pipeline.；Gemma terms are not an OSI open-source licence.
当前建议：可复现权重候选；下载前固定revision；先用官方检查脚本确认关闭触觉通道时与基座策略的等价性。
重点标签：视觉触觉VLA基础权重、Gemma条款、需后训练、待真机复核
核验来源：来源 1 · 来源 2
- 代码 · 已核验官方资源
审核码：verified_official; 官方关系：verified_substantive_official_repository_fixed_commit; 最近核验：2026-08-14代码：released: model package, post-training launcher, data conversion/normalization, inference servers and deployment documentation
权重：released separately as three public, nongated Hugging Face revisions
数据：NeoData and the large-scale pretraining pipeline are not released
许可证：CC-BY-SA-4.0；Repository-originating material only. NOTICE identifies third-party components with their own terms; this licence does not cover the model checkpoints or undisclosed NeoData.
依赖：Python 3.11、CUDA-capable GPU、PyTorch/requirements.txt environment、manual transformers replacement patch、DINOv2 cache、matching canonical 32-dimension action/state schema、websocket real-robot or ZMQ simulator endpoint
硬件/传感器：CUDA-capable workstation for post-training/inference、documented physical execution requires an embodiment-specific client, calibrated tactile baseline and robot safety/control stack、paper author platforms: Flexiv, PiPER and TacUMI; no humanoid or wheeled-leg deployment package verified
实机证据：The paper/project report author NeoReal results across nine contact-rich tasks, but task-level trial denominators and raw logs are not released.
独立复现：No qualifying independent fixed-version physical reproduction was located.
限制/反面证据：The released source explicitly excludes the large-scale pretraining pipeline.；NeoData is described but not downloadable, so training-data coverage, consent and licence cannot be independently audited.；The repository requires a manual transformers patch and a tactile-baseline/calibration convention; no container or end-to-end hardware safety closure is verified.；No independent physical reproduction and no humanoid/wheeled-leg execution package.
当前建议：视觉触觉VLA主流技术2.0·可后训练与推理复核；可将官方固定代码和特定任务checkpoint用于受控后训练/仿真验证；真机前应先验证动作32维映射、零接触基线、触觉传感器时序、网络控制延迟和安全限幅。
重点标签：视觉触觉VLA、预测触觉token、主流技术迭代、近期热门、代码已核验、权重已核验、数据未开放、待独立复现
核验来源：来源 1 · 来源 2 · 来源 3
- 权重 · 已核验官方资源
审核码：verified_official; 官方关系：verified_public_nongated_official_huggingface_revision; 最近核验：2026-08-14代码：released separately
权重：public nongated NeoSim dual-arm task checkpoint
数据：training data not released
许可证：LicenseRef-Gemma-Terms；Checkpoint inherits Gemma terms; repository code licence does not apply to weights.
依赖：N0-VTLA source、NeoSim simulator/ZMQ path、sim_dual_arm_tactile configuration
硬件/传感器：CUDA-capable GPU for inference/simulation
实机证据：This released checkpoint is documented for NeoSim dual-arm simulation, not a real robot.
独立复现：No qualifying independent replay was located.
限制/反面证据：Simulation-specific checkpoint; do not represent as real-hardware evidence.；Training data and complete simulator asset provenance remain unverified.
当前建议：仿真最小验证入口；适合先验证双臂输入/输出接口及系统延迟；真机迁移仍需独立校准与安全验证。
重点标签：NeoSim、双臂视觉触觉VLA、仿真checkpoint、Gemma条款
核验来源：来源 1 · 来源 2
- 权重 · 已核验官方资源
审核码：verified_official; 官方关系：verified_public_nongated_official_huggingface_revision; 最近核验：2026-08-14代码：released separately
权重：public nongated UniVTAC single-arm task checkpoint
数据：training data not released
许可证：LicenseRef-Gemma-Terms；Checkpoint inherits Gemma terms; repository code licence does not apply to weights.
依赖：N0-VTLA source、UniVTAC simulator/ZMQ path、sim_single_arm_tactile configuration
硬件/传感器：CUDA-capable GPU for inference/simulation
实机证据：This released checkpoint is documented for UniVTAC simulation, not a real robot.
独立复现：No qualifying independent replay was located.
限制/反面证据：Simulation-specific checkpoint; do not represent as real-hardware evidence.；Training data and complete simulator asset provenance remain unverified.
当前建议：仿真最小验证入口；适合先验证依赖、32维动作映射与触觉开关，不适合直接外推真机。
重点标签：UniVTAC、视觉触觉VLA、仿真checkpoint、Gemma条款
核验来源：来源 1 · 来源 2
N0-VTLA结合预测触觉token与ALTER离线优势标签，九项NeoReal均值47.2%对29.4%内部π0.5基线；主报告未给各任务分母和训练算力，官方仓当前只有README与非商用许可证，代码、权重和NeoData均未兑现。