ACTION Robbyant's LingBot-VLA 2.0 is a 6B Vision-Language-Action foundation model
ACTION Robbyant's LingBot-VLA 2.0 is a 6B Vision- ...
LingBot-VLA 2.0 is a 6B parameter vision-language-action foundation model developed by Robbyant. It connects visual understanding, language instructions, and physical robot actions across different robot embodiments. The model is trained on 60,000 hours of data, including 50,000 hours of robot trajectories and 10,000 hours of egocentric human videos. It extends action representation to include mobile-base movements and uses future-aware video and depth data for improved temporal reasoning. Evaluated on real-world and simulated tasks, it is described as research software rather than a standalone robot.
Source: Robotics — Robot learning news discovery · Read original article ↗
Article text · Original source · English
Log In
AI Cyborg 2.0's Post
- * *
[](https://www.facebook.com/AICyborg2?__cft__[0]=AZj-cmWIDiOQb3rFCoG-d7zVO8mPDQk57l2bS7k6QaOxOTAITK1G2H4LzFukOecuQkc-2o8zUWBYGFE0_p_V-FepYMRqSEUSnHViO4g8xc984cRhLvodqec_vF3IU9O1bMnfI0KI9HHb5_O6YaBmY4-MZQZTrpQLeKrI&__tn__=%3C%3C%2CP-R)
AI Cyborg 2.0
AI content
· 2 days ago2d ·
LingBot-VLA 2.0 | VISION • LANGUAGE • ACTION
Robbyant’s LingBot-VLA 2.0 is a 6B Vision-Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments.
Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to include hands, head, waist and mobile-base movements.
The system also uses future-aware video and depth information to improve temporal reasoning and scene understanding. Robbyant evaluates it on real-world manipulation, long-horizon mobile manipulation and RoboTwin 2.0 simulation tasks.
Released in 2026, LingBot-VLA 2.0 is research software and a foundation model rather than a standalone humanoid robot. 
Source: Robbyant, LingBot-VLA 2.0 Technical Report, 2026
#LingBotVLA#PhysicalAI#EmbodiedAI#RoboticsAI#Robotics
All reactions:
6
6 shares
Like
Comment
See more from AI Cyborg 2.0
See more from AI Cyborg 2.0
Email or phone number
Password
Log In
or
Create new account
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
robot trajectories
≈50,000 hours
View original evidence
Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments. Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to
Open source S3egocentric human videos
10,000 hours
View original evidence
Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments. Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to
Open source S3data hours
≈60,000 hours
View original evidence
Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments. Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to
Open source S3Robot count
20 robots
View original evidence
Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments. Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to
Open source S3
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Language-Action foundation model built to connect visual understanding, language instructions and physical robot actions across different robot embodiments. Its pretraining corpus contains around 60,000 hours of data: approximately 50,000 hours of robot trajectories covering 20 robot configurations plus 10,000 hours of egocentric human videos. The unified action representation extends beyond dual-arm manipulation to
Open source S3
include hands, head, waist and mobile-base movements. The system also uses future-aware video and depth information to improve temporal reasoning and scene understanding. Robbyant evaluates it on real-world manipulation, long-horizon mobile manipulation and RoboTwin 2.0 simulation tasks. Released in 2026, LingBot-VLA 2.0 is research software and a foundation model rather than a standalone humanoid robot. Source: Robb
Open source S4
Source：Robotics — Robot learning news discovery · facebook.com