Ant Group Opens-Source LingBot-Touch to Translate Heterogeneous Tactile Signals for Robotic Manipulation
Token卷到触觉了！机器人终于有了通用手感
Ant Group's LingBot team has open-sourced LingBot-Touch, a tactile pre-trained foundation model that translates heterogeneous sensor signals from various fingertips into unified tokens. The model addresses the challenge of raw tactile data being sparse and noisy by mapping different sensor layouts to a standardized fingertip space and predicting contact representations in feature space rather than reconstructing raw readings. Benchmarks show that integrating LingBot-Touch with action models like π0.5 significantly improves task success rates on unseen hardware (e.g., XHand) compared to using raw signals or no tactile input, demonstrating its ability to generalize across different robotic hands and sensor configurations.
Source: QbitAI — Robotics · 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
4,983 hours
frames
≈1,330,000,000 frames
Reported success rate
39%
Reported trials
Reported success rate
15%
Reported trials
Reported success rate
57%
Reported trials
Reported success rate
48%
Reported trials
Reported success rate
49.71%
Reported trials
Reported success rate
48%
Reported trials
- paper: 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.
其中三分之二来自人戴着触觉手套干活，剩下三分之一来自遥操作Sharpa Wave双臂机器人。
Open source S20
LingBot-Touch的预训练数据一共有 4983小时 、约 13.3亿帧 。
Open source S19
他们选用的动作模型是 π0.5 ，在只靠摄像头的情况下，成功率来到了 39% 。
Open source S5
AUV（哎呦喂）……成功率直接 掉到了15%！
Open source S7
同样是Sharpa Wave灵巧手和π0.5这个组合，接上原始触觉后去做 交接盒子 的任务，成功率从57%掉到48%。
Open source S8
即便是把灵巧手换成XHand，6项任务平均下来，成功率也从49.71%降到了48%。
Open source S9
Source：QbitAI — Robotics · qbitai.com