Robbyant open-sources tactile foundation model LingBot-Touch

Nov 10

1. 16:36

   Shanghai Observer

   Ant Lingbo Open-Sources Tactile Foundation Model LingBot-Touch

   Ant Lingbo Technology has open-sourced the tactile foundation model LingBot-Touch, pretrained on nearly 5,000 hours of real tactile data, covering three types of sensing principles: magnetic tactile, visuotactile, and piezoresistive, and outputting unified tactile Tokens for use by different dexterous hands and action models. In 26 real-robot tests on two dexterous hands, Sharpa Wave and XHand1, after integration, the average task success rates of four action models—LingBot-VLA, π0.5, Diffusion Policy, and T-Rex—increased by approximately 14 to 26 percentage points compared with when no tactile sensing was used. The technical report, model, and code have been open-sourced on Hugging Face, GitHub, and the ModelScope community.

   Expand summary
2. 16:36

   Shangguan NewsRelated discussion

   Shangguan News joined this discussion ↗

   This post contributes to heat; it is not a formal report or confirmed fact.

3. 11:38

   Phoenix Technology

   Ant Lingbo Open-Sources Tactile Foundation Model LingBot-Touch, 5.32M Parameters Integrates with Mainstream Action Models

   Ant Lingbo Technology released and open-sourced the general-purpose tactile foundation model LingBot-Touch, which uniformly converts three types of tactile signals—vision-based tactile sensing, magnetic tactile sensing, and piezoresistive tactile sensing—into general tactile Tokens usable by action models. The team used human tactile gloves plus real-robot teleoperation to collect nearly 5,000 hours and 1.33 billion frames of real tactile data, two-thirds of which came from human gloves. After integration with T-Rex, π0.5, and LingBot-VLA, the average success rates improved by 23.18, 21.81, and 24.81 percentage points, respectively. The lightest Tiny version has only 5.32 million parameters, and after integration the parameter count increases by only 0.13% to 0.16%.

   Expand summary
4. 10:49

   English text is being prepared.

   Ant Lingbo Open-Sources Tactile Foundation Model LingBot-Touch

   Ant Lingbo Technology has open-sourced the tactile foundation model LingBot-Touch, pretrained on nearly 5,000 hours of real tactile data, covering three types of sensing principles—magnetic tactile, visuotactile, and piezoresistive—and capable of converting different signals into a unified tactile representation. In 26 real-robot tests on two dexterous hands, the Sharpa Wave and XHand, after integrating this representation, the average task success rates of four action models—LingBot-VLA, π0.5, Diffusion Policy, and T-Rex—increased by approximately 14 to 26 percentage points compared with when no tactile sensing was used. All 26 tests used the lightweight version LingBot-Touch-Tiny with approximately 5.3 million parameters; after integration, the model parameter count increased by only about 0.13% to 0.16%.

   Expand summary
5. 10:49

   IT Home · AIRelated discussion

   IT Home · AI joined this discussion ↗

   This post contributes to heat; it is not a formal report or confirmed fact.

6. 07:27

   amitRelated discussion

   amit joined this discussion ↗

   This post contributes to heat; it is not a formal report or confirmed fact.

7. 04:40

   Adina YakupRelated discussion

   Adina Yakup joined this discussion ↗

   This post contributes to heat; it is not a formal report or confirmed fact.

8. 03:11

   Robbyant Ant Lingbo

   Ant Robbyant Open-Sources Tactile Foundation Model LingBot-Touch

   Ant Robbyant has open-sourced the tactile foundation model LingBot-Touch, aimed at dexterous manipulation and pretrained on nearly 5,000 hours of real tactile data. Through unified heterogeneous tactile modeling, the model supports transfer across sensors, dexterous hands, tasks, and action models.

   Expand summary