Robbyant, an embodied artificial intelligence company under Ant Group, released an open-source tactile AI model called LingBot-Touch on November 10, 2026. The model helps robots understand and process touch signals from different sensor types. It was pretrained on nearly 5,000 hours of real-world tactile data.
Sensor compatibility
LingBot-Touch processes data from magnetic, visual, and piezoresistive sensors. It uses a shared representation and a Haptic Transformer to learn unified contact physics. The design aims to make tactile technology work across different robotic systems and sensor layouts.
Real-world testing
Robbyant evaluated the model through 26 tests involving two robotic hands and four action models. Average task success rates improved by around 14 to 26 percentage points compared with systems without tactile input. The benchmark tasks covered tactile-guided fine manipulation, fragile and deformable object handling, and bimanual coordination.
Open-source robotics effort
The release is part of Ant Group's wider open-source robotics efforts. Robbyant is working toward a shared brain for all robots. The company aims to support robots that operate in real-world environments rather than laboratory demonstrations.
The smallest encoder configuration has 5.46 million parameters and supports high-frequency inference. The model also transfers to unseen tactile sensors and real-world dexterous robot hands. Robbyant reports continued performance gains as tactile training data scales.