Skip to content
Chinese robotics — Dataset and collection discovery·· 3 days agoSignalEditorial score75

Ant Group's Lingbo Opens-Source Tactile Foundation Model LingBot-Touch for Robot Dexterous Manipulation

蚂蚁灵波开源触觉基础模型LingBot‑Touch,为机器人通用大脑补上指尖 “手感”

Summary

On November 10, Ant Group's Lingbo officially open-sourced the tactile foundation model LingBot-Touch to address gaps in robot tactile perception. Pretrained on nearly 5,000 hours of real-world tactile data, the model standardizes signals from magnetic, visual-tactile, and piezoresistive sensors into a unified representation. Tests on Sharpa Wave and XHand robots across 26 tasks showed that integrating LingBot-Touch improved average task success rates by 14–26 percentage points compared to non-tactile approaches.

Source: Chinese robotics — Dataset and collection discovery · Read original article ↗

English translation unavailable. Choose the original article to read it.

What the source reports

Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.

Reported numbers

  • training_hours

    5,000 hours

    View original evidence
    据了解,LingBot‑Touch 基于接近 5000 小时的真实触觉数据完成预训练,兼容磁触觉、视触觉、压阻式三类主流触觉传感方案,能够把异构传感器输出的差异化原始信号,转化为一套标准化触觉表征,实现触觉经验跨传感器、跨灵巧手、跨算法模型、跨任务场景复用。
    Open source S3
  • training_hours

    5,000 hours

    View original evidence
    全部 26 项真机实验均采用参数量约 530 万的 LingBot‑Touch‑Tiny 轻量版,接入现有动作模型后,整体参数量仅增加 0.13%‑0.16%,适配端侧部署需求;更大规模的 LingBot‑Touch‑Small 版本,在感知任务上具备性能提升潜力,在接近 5000 小时预训练数据规模下,模型依旧能够从新增数据中持续获益,也预示触觉基础模型还有较大迭代空间。
    Open source S20
  • model_parameters

    ≈5,300,000

    View original evidence
    全部 26 项真机实验均采用参数量约 530 万的 LingBot‑Touch‑Tiny 轻量版,接入现有动作模型后,整体参数量仅增加 0.13%‑0.16%,适配端侧部署需求;更大规模的 LingBot‑Touch‑Small 版本,在感知任务上具备性能提升潜力,在接近 5000 小时预训练数据规模下,模型依旧能够从新增数据中持续获益,也预示触觉基础模型还有较大迭代空间。
    Open source S20
  • Reported success rate

    56.25%

    Reported trials

    View original evidence
    其中拧瓶盖测试中,T‑Rex 控制 Sharpa Wave 灵巧手,无触觉条件下成功率仅 56.25%;直接输入传感器原始信号时成功率为 62.5%;接入 LingBot‑Touch 触觉表征后,任务成功率提升至 90%,直观证明标准化触觉信息对机器人实操能力的增益。
    Open source S11
  • Reported success rate

    62.5%

    Reported trials

    View original evidence
    其中拧瓶盖测试中,T‑Rex 控制 Sharpa Wave 灵巧手,无触觉条件下成功率仅 56.25%;直接输入传感器原始信号时成功率为 62.5%;接入 LingBot‑Touch 触觉表征后,任务成功率提升至 90%,直观证明标准化触觉信息对机器人实操能力的增益。
    Open source S11
  • Reported success rate

    90%

    Reported trials

    View original evidence
    其中拧瓶盖测试中,T‑Rex 控制 Sharpa Wave 灵巧手,无触觉条件下成功率仅 56.25%;直接输入传感器原始信号时成功率为 62.5%;接入 LingBot‑Touch 触觉表征后,任务成功率提升至 90%,直观证明标准化触觉信息对机器人实操能力的增益。
    Open source S11
  • paper: not_reported
  • code: not_reported
  • weights: not_reported
  • dataset: not_reported

What remains unknown

Not established in the collected evidence: Environment, Control, Data origin.

Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.

Source excerpts and review record

Automatically extracted; no manual editorial approval recorded.

据了解,LingBot‑Touch 基于接近 5000 小时的真实触觉数据完成预训练,兼容磁触觉、视触觉、压阻式三类主流触觉传感方案,能够把异构传感器输出的差异化原始信号,转化为一套标准化触觉表征,实现触觉经验跨传感器、跨灵巧手、跨算法模型、跨任务场景复用。

Open source S3

全部 26 项真机实验均采用参数量约 530 万的 LingBot‑Touch‑Tiny 轻量版,接入现有动作模型后,整体参数量仅增加 0.13%‑0.16%,适配端侧部署需求;更大规模的 LingBot‑Touch‑Small 版本,在感知任务上具备性能提升潜力,在接近 5000 小时预训练数据规模下,模型依旧能够从新增数据中持续获益,也预示触觉基础模型还有较大迭代空间。

Open source S20

其中拧瓶盖测试中,T‑Rex 控制 Sharpa Wave 灵巧手,无触觉条件下成功率仅 56.25%;直接输入传感器原始信号时成功率为 62.5%;接入 LingBot‑Touch 触觉表征后,任务成功率提升至 90%,直观证明标准化触觉信息对机器人实操能力的增益。

Open source S11

Source:Chinese robotics — Dataset and collection discovery · donews.com

Timezone · UTC

Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).