code_weights_data_release_state · verified_hold
审核码：verified_hold; 官方关系：verified_official_placeholder_method_code_promised_not_released; 最近核验：2026-07-23代码：Fixed commit 2375a73bd26eef284402c2c25107be9b1905b100 contains README.md, a preview GIF and an MP4 demo only; README states that code will be released soon.
权重：No Force-VAE, action-expert or FM-VLA checkpoint is released.
数据：The 650 demonstrations across the three tasks and force/state logs are not released.
许可证：not_disclosed；No repository LICENSE exists. The project-page template's CC BY-SA footer applies to the page/template and cannot license absent method code, weights or demonstration data.
依赖：π0.5 VLA base、Perceiver-IO force VAE、six-axis force/torque sensing at 100 Hz、three RGB streams and AgiBot G1 control stack
硬件/传感器：AgiBot G1 with two 7-DoF arms and grippers、wrist six-axis force/torque sensors、paper inference benchmark uses an NVIDIA RTX 4090
实机证据：The project reports 18 trials per task and 83.3% average FM-VLA success over cups, buttons and wiping.；The placeholder repository contains author demo media but no executable evaluation or robot-control path.
独立复现：No independent code, checkpoint, 650-demonstration corpus, force-log replay or physical task reproduction is public.
限制/反面证据：No method code despite the repository's Code label and official project link.；No checkpoints, demonstration data, force logs, configuration or dependency lock.；No artifact license.；No independent reproduction.
当前建议：高潜待持续验证；Keep FM-VLA high on the release watchlist because the physical evidence and latency gain are strong, but do not expose a code badge until executable method files and a license appear.
重点标签：接触记忆VLA颠覆性分支候选、artifact:official-placeholder、code:coming-soon、community:unverified
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4