aifollow.news
← TrendingLatest development Nov 10 10:46
STORY IN FOCUS

Robbyant releases open-source robot touch model LingBot-Touch

LingBot-Touch turns fingertip sensor readings into signals robot control software can use and is licensed under Apache 2.0. The cited account says its JEPA version reached 49.3% top-5 accuracy on tactile-to-video retrieval after training on 4,000 hours of data, with model size and touch data held the same across the compared encoders. The reported result is not an independent verification.

1 developments5 reports3 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

Ant LingBo open-sources LingBot-Touch for reusable robotic touch perception

Ant LingBo Technology has open-sourced LingBot-Touch, a tactile foundation model that maps signals from three sensor types into a shared representation for different dexterous hands and action models. According to its official description, across 26 physical-robot tests on two dexterous hands, four action models saw average task success rates rise by about 14 to 26 percentage points compared with using no touch input.

View source reports · 3 sources · 5 reports
All reports · 5

By original publication time, with each report's bookmarks and feedback preserved.

Rohan Paul (@rohanpaul_ai) Machine translation

Robbyant releases open-source robot touch model LingBot-Touch

LingBot-Touch turns fingertip sensor readings into signals robot control software can use and is licensed under Apache 2.0. The cited account says its JEPA version reached 49.3% top-5 accuracy on tactile-to-video retrieval after training on 4,000 hours of data, with model size and touch data held the same across the compared encoders. The reported result is not an independent verification.

Rohan Paul (@rohanpaul_ai) Machine translation

Robbyant releases open-source robot touch model LingBot-Touch

LingBot-Touch turns fingertip sensor readings into signals robot control software can use and is released under Apache 2.0. The source reports 49.3% top-5 accuracy for its JEPA version on tactile-to-video retrieval with the same touch data and model size; adding data from other sensors improved seven of eight sensor-task pairs.

量子位 AI Machine translation

蚂蚁灵波 releases LingBot-Touch tactile model for different robotic hands and action models

LingBot-Touch converts magnetic, vision-based and piezoresistive sensor signals into a shared tactile representation. Its technical report, model and code are available to download. 蚂蚁灵波 says that across 26 physical-robot tests on two dexterous hands, four action models improved their average task success rates by about 14 to 26 percentage points over runs without touch input. The results come from the company's own tests.

量子位 AI Machine translation

Ant Group’s LingBot team open-sources the LingBot-Touch robot tactile model

LingBot-Touch encodes readings from three types of tactile sensors in a shared format for use by different action models; its technical report, code and model weights are available. In the team’s reported robot tests, π0.5 averaged a 74.17% success rate across six tasks with XHand, whose sensor model was absent from pretraining data, versus 48.00% when fed raw tactile signals.

IT之家 科技新闻 Machine translation

Ant LingBo open-sources LingBot-Touch for reusable robotic touch perception

Ant LingBo Technology has open-sourced LingBot-Touch, a tactile foundation model that maps signals from three sensor types into a shared representation for different dexterous hands and action models. According to its official description, across 26 physical-robot tests on two dexterous hands, four action models saw average task success rates rise by about 14 to 26 percentage points compared with using no touch input.