LingBot-VLA 2.0
#7 en Automatisation IA & workflowsOpenroboto · v2.0 · depuis Juli 2026 (angekündigt am 8. Juli 2026) · 3× · vu le 07 sept. 2026
LingBot-VLA 2.0 is an open-source Vision-Language-Action foundation model from Robbyant, an embodied AI company within Ant Group, designed to control robots across diverse morphologies. The model was pretrained on approximately 60,000 hours of data (50,000 hours of robot trajectories across 20 robot configurations plus 10,000 hours of egocentric human video data) and supports arms, grippers, hands, head, waist, and mobile base control. It is built on a Qwen3-VL-4B-Instruct vision-language backbone with a Mixture-of-Transformers/MoE architecture, achieving inference latency under 130ms on an RTX 4090. Code and model weights (6B checkpoint) are freely available under an Apache 2.0 license on GitHub and Hugging Face.
Fonctionnalités
| Deployment Model | Self-hosted / local inference, e.g. on RTX 4090 (<130ms latency) |
| Use Case Scope | Robot control: retail sorting, logistics, industrial automation, care/household robots |
| Integrations | 20 robot configurations, 17 hardware brands (incl. Unitree, AgiBot, AgileX, Galaxea, Astribot, Franka) |
| License | Apache 2.0 (codebase) |
| Platform | GitHub & Hugging Face (code, technical report, 6B checkpoint 'lingbot-vla-v2-6b') |
| Price | Free (open source) |
| Release Date | July 8, 2026 (successor to LingBot-VLA 1.0, released January 2026) |