Ant Lingbo Opens Tactile Foundation Model LingBot-Touch to Accelerate Embodied AI
Fable 5.2 “撞车”蚂蚁灵波触觉模型，CEO 朱兴：两类模型协同将加速具身产业成熟
On November 10, Ant Lingbo Technology open-sourced the tactile foundation model LingBot-Touch alongside the release of Claude Fable 5.2. Pretrained on nearly 5,000 hours of real-world touch data covering magnetic, visual, and resistive sensing principles, the model unifies diverse sensor signals into a common representation for cross-hardware reuse. Tests using the lightweight LingBot-Touch-Tiny variant (approx. 5.3M parameters) on Sharpa Wave and XHand1 showed task success rates for four action models improved by 14 to 26 percentage points compared to no touch input.
Source: Chinese robotics — Robot learning research · 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
此次开源的 LingBot-Touch 基于近 5000 小时真实触觉数据预训练，覆盖磁触觉、视触觉和压阻式三类主流传感原理，可将不同信号转化为同一套触觉表征，供不同 灵巧手 和动作模型使用，让机器人在"看得清楚、想得明白"的同时，进一步补齐精细操作所需的触觉感知。
Open source S7model_parameters
≈5,300,000
View original evidence
为兼顾感知能力与端侧算力，蚂蚁灵波提供了两种容量的模型：此次 26 项真机测试全部使用约 530 万参数的轻量版 LingBot-Touch-Tiny 完成，接入 LingBot-VLA、π0.5 和 T-Rex 后，模型参数量仅增加约 0.13% 至 0.16%；容量较大的 LingBot-Touch-Small 则在感知测试中表现出进一步提升的潜力——当预训练数据增至近 5000 小时时，模型仍能从新增数据中获益，说明触觉预训练还有持续扩展提升的空间。
Open source S19Reported success rate
56.25%
Reported trials
View original evidence
以 T-Rex 控制 Sharpa Wave 拧瓶盖为例，不使用触觉时任务成功率为 56.25%，直接输入传感器原始信号为 62.5%，接入 LingBot-Touch 触觉表征后成功率达到 90%。
Open source S13Reported success rate
90%
Reported trials
View original evidence
以 T-Rex 控制 Sharpa Wave 拧瓶盖为例，不使用触觉时任务成功率为 56.25%，直接输入传感器原始信号为 62.5%，接入 LingBot-Touch 触觉表征后成功率达到 90%。
Open source S13
Dataset details reported by the source
- Embodiments
- XHand1 Open source S18
- 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.
Dataset licensing and commercial-use rights are not established here.
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.
此次开源的 LingBot-Touch 基于近 5000 小时真实触觉数据预训练，覆盖磁触觉、视触觉和压阻式三类主流传感原理，可将不同信号转化为同一套触觉表征，供不同 灵巧手 和动作模型使用，让机器人在"看得清楚、想得明白"的同时，进一步补齐精细操作所需的触觉感知。
Open source S7
由此，人戴触觉手套和遥操作灵巧手采集的数据可以共同用于训练，不同动作模型也可以直接调用训练好的模型提取接触特征，这一架构还为未来接入新型触觉传感器留出了扩展空间。
Open source S17
（图说：LingBot-Touch 将磁触觉、视触觉、压阻式触觉信号编码为统一的触觉 Token，供不同动作模型使用） 在 XHand1 上，团队进一步验证了触觉能力的跨硬件复用：XHand1 采用的磁触觉传感器型号并不在预训练数据中，模型学到的磁触觉经验全部来自人戴手套采集的数据。
Open source S18
为兼顾感知能力与端侧算力，蚂蚁灵波提供了两种容量的模型：此次 26 项真机测试全部使用约 530 万参数的轻量版 LingBot-Touch-Tiny 完成，接入 LingBot-VLA、π0.5 和 T-Rex 后，模型参数量仅增加约 0.13% 至 0.16%；容量较大的 LingBot-Touch-Small 则在感知测试中表现出进一步提升的潜力——当预训练数据增至近 5000 小时时，模型仍能从新增数据中获益，说明触觉预训练还有持续扩展提升的空间。
Open source S19
26 项真机测试，任务成功率提高 14 至 26 个百分点 随着通用大模型持续进步，机器人理解复杂任务、推理操作步骤的能力正在增强。
Open source S9
以 T-Rex 控制 Sharpa Wave 拧瓶盖为例，不使用触觉时任务成功率为 56.25%，直接输入传感器原始信号为 62.5%，接入 LingBot-Touch 触觉表征后成功率达到 90%。
Open source S13
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Check the exact dataset version, license, modalities and task coverage against the intended use.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：Chinese robotics — Robot learning research · news.qq.com