We're open-sourcing LingBot-VA to accelerate the development of embodied intelligence for the entire community. Let's build the future of robotics together. Check out the code, models, and our tech report: 🐙 Code: https://t.co/whqNyDZEN5 🤗 Hugging Face:

🦾 From seeing to doing. We're closing the loop between video prediction and real-world action. On the final day of Robbyant Open Source Week, we bring you LingBot-VA—the world's first causal video-action world model for generalist robot control. 🔥 Key Highlights: 🤖 Predicts &

🔥 Very excited to share that we’re releasing LingBot-World 🌍 @robbyant_brain — an open-source frontier world model! We’re pushing the limits of: 🔹 High-Fidelity Simulation & Precise Control 🔹 Long-Horizon Consistency & Memory 🔹 Modeling Physical & Game Worlds The most

Thanks to AK for the rec! 👏 Dig into LingBot-VLA’s tech details → our technical report is up! 📄 #LingBotVLA #TechReport #VLA

From perception (LingBot-Depth) to action (LingBot-VLA) to imagination (LingBot-World), we are building the foundational stack for embodied intelligence. Day 3 of our open-source week. Dive in: 🌐 Website: https://t.co/YP48j9M0S1 📑 Tech Report: https://t.co/t6ueL9dDH9 🐙 Code:

Zero-shot generalization: feed LingBot-World a single real-world photo or game screenshot, and it generates a fully interactive world—no scene-specific training needed. This is powered by our hybrid data strategy: large-scale web videos + game captures with clean, UI-free frames

A true training ground must respond in real-time. LingBot-World achieves ~16 FPS throughput with under 1-second end-to-end latency. Control characters, adjust camera angles, or trigger environment changes via text commands—all with instant visual feedback. No pre-rendering, just

Long-term consistency is a known challenge for video generation—objects distort, scenes collapse. LingBot-World solves this with multi-stage training and parallel acceleration, enabling nearly 10 minutes of stable, continuous generation. Even after the camera looks away for 60

🌍 Reality is expensive. Simulation is the shortcut. But what if the simulation could think, respond, and remember? Today, we open-source LingBot-World, an interactive world model built on @Alibaba_Wan Wan2.2! 🔥 We’re pushing the limits of: 🔷 High-Fidelity Simulation &

🚀 Meet LingBot-VLA: A pragmatic Vision-Language-Action model designed to bridge the gap between perception and execution in robotics. 🤖 ✅LingBot-VLA-4B: Lightweight & versatile. https://t.co/eXdtpMhNfo ✅LingBot-VLA-4B-Depth: Enhanced for high-precision spatial tasks.

Thanks for the shout-out, @AdinaYakup! We're excited to share our LingBot-VLA and LingBot-Depth models with the community. Check out the technical reports and explore the models on Hugging Face. Stay tuned—something even more powerful is coming soon!

From LingBot-Depth to LingBot-VLA, we are building the foundational blocks for embodied intelligence. This is day 2 of our open-source week—stay tuned for more. Dive in and build with us: 🌐Website: https://t.co/SgNKPmWbKZ 📊 Datasets: https://t.co/mX8A919eS7 🐙 Code:

Lowering the barrier to entry is critical. We're also open-sourcing our entire post-training toolchain. It's 1.5-2.8x more efficient than mainstream frameworks like StarVLA, enabling developers to fine-tune on their own tasks with significantly less data and compute.

We also integrated the spatial awareness of LingBot-Depth, which we released yesterday. By distilling depth information into our VLA via learnable queries, the average success rate on GM-100 climbs further to 17.3%. Combining action models with high-fidelity perception is a key

Performance across real and simulated environments: GM-100 real-robot benchmark (100 tasks, 3 robot embodiments): LingBot-VLA hits a 15.7% cross-embodiment success rate → outperforming Pi0.5 (13.0%). RoboTwin 2.0 (heavily randomized simulation): We hold a 9.92% lead in

🧠 What if one AI brain powers all robots? Retraining for every new embodiment is the biggest scaling pain in embodied AI—we’re fixing it. Today, we open-source LingBot-VLA: a Vision-Language-Action model built on @Alibaba_Qwen Qwen-2.5-VL and pre-trained on 20,000 hours of

Very excited to share our first public release after I joined @robbyant_brain! We present Lingbot-Depth 👀 — a state-of-the-art depth foundation model trained with RGB-D MAE on millions of real & simulated RGBD pairs. 🔹 Camera depths as natural masks for RGB-D MAE modeling 🔹

@robbyant_brain has open-sourced LingBot-Depth, a spatial intelligence model that lets robots "see" the unseeable. 🚀By aligning RGB & Depth latent spaces, it achieves reliable grasping of transparent/reflective objects where traditional sensors fail. Key Breakthroughs: ✅SOTA

We believe in building in the open. Today, we're releasing it all: 🌐Website: https://t.co/HX1KOigCqq 🐙Code: https://t.co/0cVHhGyoqy 📑Tech Report: https://t.co/O88ficd6oS 🤗HuggingFace: https://t.co/Dk0nlLgFCD 🤖ModelScope: https://t.co/tRvawRXVtc And we're not stopping. Stay

This isn't just about grasping. LingBot-Depth provides a robust foundation for a wide range of spatial intelligence tasks. This includes more accurate 3D indoor mapping, improved camera pose and trajectory estimation, and reliable 4D point tracking of dynamic objects—all within a