Robbyant releases LingBot-VLA 2.0 universal AI brain for robots

Robbyant, an embodied AI company within Ant Group, has announced the open-source release of LingBot-VLA 2.0, the successor to LingBot-VLA 1.0, which was released in January 2026. The new model is positioned as an advancement in vision-language-action (VLA) technology—AI systems that understand visual input, language instructions, and generate robotic actions. The upgrade addresses what the company identifies as the primary bottleneck in industrial-scale robotics deployment: the lack of a truly universal AI brain capable of controlling robots across different morphologies and configurations.

LingBot-VLA 2.0 was pre-trained on 60,000 hours of high-quality, real-world physical data. This training dataset was carefully curated from two sources: 50,000 hours of cleaned real-robot interaction data and 10,000 hours of distilled first-person human manipulation data. Crucially, this data was sourced from 20 distinct robot morphologies across 17 leading manufacturers—Leju, AgiBot, Unitree, AgileX, Galaxea, Galbot, Astribot, RealMan, Franka, ARX, X-Humanoid, Fourier, MagicLab, Spirit AI, Zerith, Flexiv, and Qinglong—covering single-arm, dual-arm, bipedal, and wheeled robot configurations. The model expands its degrees-of-freedom (DoF) support to include not only traditional arm control but also head, waist, end-effectors (hands), and mobile chassis, enabling coordinated whole-body control across diverse robot architectures.

Performance testing demonstrates measurable improvements over competing models. On the Shanghai Jiao Tong University's GM-100 benchmark for dual-arm manipulation, LingBot-VLA 2.0 achieved leading average task progress scores and success rates on both AgileX Cobot Magic and Galaxea R1 Pro platforms, outperforming both π0.5 and GR00T N1.7. In long-horizon mobile manipulation tasks tested on the ARX Arm + AgileX Chassis and Astribot S1 platforms, LingBot-VLA 2.0 surpassed π0.5 in both task progress and success rates. These results highlight the model's advanced capability in executing long-sequence tasks and generalizing across different robot platforms.

A critical engineering achievement is the dramatic improvement in deployment efficiency. Compared to the previous generation, inference efficiency has been improved by 3 times, with latency strictly maintained under 150 milliseconds. This improvement is significant because it substantially lowers the barrier for real-time commercial applications—reducing both computational cost and response time, two factors essential for widespread industrial adoption. Robbyant is actively testing LingBot-VLA 2.0 in real-world business scenarios through partnerships with hardware manufacturers Leju and Ti5Robot, and enterprise customers GuoDa Drugstore and Longsheng Technology, conducting comprehensive commercial pilot testing in retail sorting, logistics, and industrial environments. Additionally, Robbyant is collaborating with companies like GenRobot.ai to build standardized data ecosystems, signaling an effort to establish interoperable infrastructure for embodied AI development.

LingBot-VLA 2.0 represents a significant step toward solving a long-standing constraint in embodied robotics. The embodied AI industry has advanced hardware and control systems, but the lack of a universal AI brain capable of controlling robots across different morphologies has been a primary bottleneck for scaling deployment to industrial settings. Robbyant's approach of training on 60,000 hours of real-world physical data—drawn from 20 different robot types manufactured by 17 companies—is designed to create genuine cross-morphology generalization rather than task-specific models.

The performance improvements demonstrated on established benchmarks suggest the model delivers real capability gains. On the Shanghai Jiao Tong University's GM-100 benchmark, LingBot-VLA 2.0 outperformed both π0.5 and GR00T N1.7 on dual-arm manipulation, and surpassed π0.5 on long-horizon mobile manipulation tasks. The 3× improvement in inference efficiency while maintaining sub-150-millisecond latency directly addresses a commercial deployment barrier—real-time responsiveness at scale. This combination of generalization and efficiency may enable broader adoption of AI-controlled robots in retail, logistics, and factory settings where response time is critical.

No comments yet. Be the first to share your thoughts!