Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Introducing TraceVLA: a fully open-source Vision-Language-Action model reimagining spatial-temporal awareness: ✨ 3.5x gains on real robots, SOTA in simulation 💡 Fine-tunes on just 150K trajectories ⚡ Compact 4B model = 7B performance

Yongyuan Liang

2,953 subscribers

39,603 views • 1 year ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

11 Comments

Yongyuan Liang1 year ago

We introduce visual trace prompting:🔹Track robot's movement via point-tracking (Co-Tracker) 🔹Overlay traces on observations Model processes: 1️⃣ Original view (preserve full info) 2️⃣ View with traces as prompts A simple yet powerful technique to boost VLA's spatial understanding

Yongyuan Liang1 year ago

TraceVLA in action: Watch it excel at diverse manipulation tasks on a real WidowX-250 robot! From soft-object handling to precision pick-and-place, TraceVLA consistently outperforms OpenVLA in both in-distribution and out-of-distribution tasks.

Yongyuan Liang1 year ago

Superior simulation results: On Google’s SimplerEnv robot tasks, TraceVLA outshines OpenVLA across all metrics in both 7B and 4B versions! 🚀 20% boost in handling: ▪️ Camera changes ▪️ Distractors ▪️ Varied visual backgrounds

Yongyuan Liang1 year ago

Efficient and lightweight: 🔸 TraceVLA requires <10GB memory on 8 H100 GPUs 🔸 Adds only 0.036s per timestep A powerful VLA upgrade with minimal overhead!

Yongyuan Liang1 year ago

Available resources include: ▫️7B TraceVLA checkpoints ▫️Lightweight 4B Phi3V-OpenVLA & TraceVLA models ▫️Fine-tuned TraceVLA models 💻 Code: 🤗 Models: Try TraceVLA family models today!

Yongyuan Liang1 year ago

Check out our project page: ArXiv: Joint work with @ruijie_zheng12 @ShuaiyiH @JianfengGao0217 @haldaume3 @Andrey__Kolobov @furongh @jw2yang4ai

Yang1 year ago

Want to learn how practical AI skills and automations for your business and work? Check out our 50+ step-by-step video tutorials 100% FREE 20+ hours of Ai and Automation goodness absolutely free 🥳

Mu Cai @ Industry Job Market1 year ago

Congratulations! Really interesting work on applying visual prompts on VLA tasks!

Yongyuan Liang1 year ago

Thanks!!!

Dmytro Kuzmenko1 year ago

thank you very much for sharing, great idea and rather impressive results!

Ray | AI marketer - Social Media Assistant1 year ago

real-time engagement is key. we help brands connect with their audience 24/7, no burnout.

Related Videos

✨ Introducing 𝐎𝐩𝐞𝐧𝐕𝐋𝐀 — an open-source vision-language-action model for robotics! 👐 - SOTA generalist policy - 7B params - outperforms Octo, RT-2-X on zero-shot evals 🦾 - trained on 970k episodes from OpenX dataset 🤖 - fully open: model/code/data all online 🤗 🧵👇

✨ Introducing 𝐎𝐩𝐞𝐧𝐕𝐋𝐀 — an open-source vision-language-action model for robotics! 👐 - SOTA generalist policy - 7B params - outperforms Octo, RT-2-X on zero-shot evals 🦾 - trained on 970k episodes from OpenX dataset 🤖 - fully open: model/code/data all online 🤗 🧵👇

Moo Jin Kim

227,128 views • 2 years ago

🚀 MiniCPM enters the physical world — enabling robots to understand, remember, and act. We open-source MiniCPM-Robot, our first embodied AI model series, including: 🤖 MiniCPM-RobotManip — a 1.5B general-purpose Vision-Language-Action (VLA) model for robotic manipulation. 🐕 MiniCPM-RobotTrack — a compact model for real-world target tracking. ⚡ PhyAI — a high-performance inference framework built for embodied models. Together, they bring efficient, practical, and open embodied intelligence closer to real-world robots. ⭐ GitHub: 🤗 MiniCPM-RobotManip: 🤗 MiniCPM-RobotTrack:

🚀 MiniCPM enters the physical world — enabling robots to understand, remember, and act. We open-source MiniCPM-Robot, our first embodied AI model series, including: 🤖 MiniCPM-RobotManip — a 1.5B general-purpose Vision-Language-Action (VLA) model for robotic manipulation. 🐕 MiniCPM-RobotTrack — a compact model for real-world target tracking. ⚡ PhyAI — a high-performance inference framework built for embodied models. Together, they bring efficient, practical, and open embodied intelligence closer to real-world robots. ⭐ GitHub: 🤗 MiniCPM-RobotManip: 🤗 MiniCPM-RobotTrack:

OpenBMB

330,472 views • 1 month ago

DynamicVLA A compact 0.4B Vision-Language-Action model that finally lets robots manipulate *moving* objects in real-time, closing the perception-execution gap with Continuous Inference and Latent-aware Action Streaming.

DynamicVLA A compact 0.4B Vision-Language-Action model that finally lets robots manipulate moving objects in real-time, closing the perception-execution gap with Continuous Inference and Latent-aware Action Streaming.

DailyPapers

16,357 views • 7 months ago

Everyone is sleeping on this new OCR model! dots-ocr is a new 1.7B vision-language model that achieves SOTA performance on multilingual document parsing. - Supports 100+ languages - Works with both images and PDFs - Handles text, tables, formulas seamlessly 100% open-source.

Everyone is sleeping on this new OCR model! dots-ocr is a new 1.7B vision-language model that achieves SOTA performance on multilingual document parsing. - Supports 100+ languages - Works with both images and PDFs - Handles text, tables, formulas seamlessly 100% open-source.

Akshay 🚀

252,205 views • 1 year ago

Apple just released and open-sourced FastVLM! FastVLM is a lightning-fast vision-language model that combines rapid image and text understanding with efficient on-device performance. 100% Open Source

Apple just released and open-sourced FastVLM! FastVLM is a lightning-fast vision-language model that combines rapid image and text understanding with efficient on-device performance. 100% Open Source

Sumanth

43,704 views • 1 year ago

🚀 First step to unlocking Generalist Robots! Introducing 🤖LAPA🤖, a new SOTA open-sourced 7B VLA pretrained without using action labels. 💪SOTA VLA trained with Open X (outperforming OpenVLA on cross and multi embodiment) 😯LAPA enables learning from human videos, unlocking potential for robotic foundation model ❗Over 30x pretraining efficiency for VLA training 🤗Code and checkpoints are all open-sourced!

🚀 First step to unlocking Generalist Robots! Introducing 🤖LAPA🤖, a new SOTA open-sourced 7B VLA pretrained without using action labels. 💪SOTA VLA trained with Open X (outperforming OpenVLA on cross and multi embodiment) 😯LAPA enables learning from human videos, unlocking potential for robotic foundation model ❗Over 30x pretraining efficiency for VLA training 🤗Code and checkpoints are all open-sourced!

Seonghyeon Ye

33,296 views • 1 year ago

Molmo by Ai2 - Open source SoTA Multimodal (Vision) Language model, beating Claude 3.5 Sonnet, GPT4V and comparable to GPT4o 🔥 They release four model checkpoints: 1. MolmoE-1B, a mixture of experts model with 1B (active) 7B (total) 2. Molmo-7B-O, most open 7B model 3. Molmo-7B-D, demo model 4. Molmo-72B, best model System Architecture > Input: Multi-scale, multi-crop images generated from the original image. > Vision Encoder: OpenAI's ViT-L/14 336px CLIP model, a powerful ViT, encodes images into vision tokens. > Connector: MLP projects tokens to LLM input space, followed by pooling for dimensionality reduction. > LLM: Decoder-only Transformer, various options (OLMo, OLMoE, Qwen2, Mistral, Gemma2, Phi) with diverse scales and openness. Model Variants > Vision Encoder: Consistent ViT-L/14 CLIP model across variants. > LLM: OLMo-7B-1024, OLMoE-1B-7B-0924, Qwen2 (7B, 72B), Mistral 7B, Gemma2 9B, Phi 3 Medium, offering different capacities and openness levels. Training Strategy > Stage 1: Multimodal pre-training for caption generation with new captioning data. > Stage 2: Supervised fine-tuning on a dataset mixture, updating all parameters. > No RLHF involved, Learning rates adjusted based on component types and pre-training status. > All the weights are available on Hugging Face Hub 🤗 > Compatible with Transformers (Remote Code) Kudos Ai2 for such a brilliant and open work! 🐐 Video credits: Allen AI YT Channel

Molmo by Ai2 - Open source SoTA Multimodal (Vision) Language model, beating Claude 3.5 Sonnet, GPT4V and comparable to GPT4o 🔥 They release four model checkpoints: 1. MolmoE-1B, a mixture of experts model with 1B (active) 7B (total) 2. Molmo-7B-O, most open 7B model 3. Molmo-7B-D, demo model 4. Molmo-72B, best model System Architecture > Input: Multi-scale, multi-crop images generated from the original image. > Vision Encoder: OpenAI's ViT-L/14 336px CLIP model, a powerful ViT, encodes images into vision tokens. > Connector: MLP projects tokens to LLM input space, followed by pooling for dimensionality reduction. > LLM: Decoder-only Transformer, various options (OLMo, OLMoE, Qwen2, Mistral, Gemma2, Phi) with diverse scales and openness. Model Variants > Vision Encoder: Consistent ViT-L/14 CLIP model across variants. > LLM: OLMo-7B-1024, OLMoE-1B-7B-0924, Qwen2 (7B, 72B), Mistral 7B, Gemma2 9B, Phi 3 Medium, offering different capacities and openness levels. Training Strategy > Stage 1: Multimodal pre-training for caption generation with new captioning data. > Stage 2: Supervised fine-tuning on a dataset mixture, updating all parameters. > No RLHF involved, Learning rates adjusted based on component types and pre-training status. > All the weights are available on Hugging Face Hub 🤗 > Compatible with Transformers (Remote Code) Kudos Ai2 for such a brilliant and open work! 🐐 Video credits: Allen AI YT Channel

Vaibhav (VB) Srivastav

80,474 views • 1 year ago

No words. Just wow. LingBot-World: A playable open-source world model built on Wan2.2+Qwen3-VL-2B; - real-time interactive simulation at 720p@16fps, <1s latency; - minute-long contextual memory. - open source!

No words. Just wow. LingBot-World: A playable open-source world model built on Wan2.2+Qwen3-VL-2B; - real-time interactive simulation at 720p@16fps, <1s latency; - minute-long contextual memory. - open source!

Wildminder

36,918 views • 7 months ago

Introducing OFT—an Optimized Fine-Tuning recipe for VLAs! Fine-tuning OpenVLA w/ OFT, we see: -25-50x faster inference ⚡️ -SOTA 97.1% avg SR in LIBERO 💪 -high-freq control w/ 7B model on real bimanual robot -outperforms π₀, RDT-1B, DiT Policy, MDT, Diffusion Policy, ACT 🧵👇

Introducing OFT—an Optimized Fine-Tuning recipe for VLAs! Fine-tuning OpenVLA w/ OFT, we see: -25-50x faster inference ⚡️ -SOTA 97.1% avg SR in LIBERO 💪 -high-freq control w/ 7B model on real bimanual robot -outperforms π₀, RDT-1B, DiT Policy, MDT, Diffusion Policy, ACT 🧵👇

Moo Jin Kim

84,333 views • 1 year ago

Humanoid robots shouldn't just follow pre-defined movements—they should perform! 🤖✨ Introducing UniAct: A unified model for multimodal motion generation and action streaming. Most humanoids are limited to pre-designed moves. UniAct changes the game by allowing robots to generate live action sequences from: 📝 Text instructions 🎶 Music rhythms 📉 Spatial trajectories 🔄 Cross-modal signals Whether it’s dancing to a beat or following a complex path, UniAct brings humanoid robots to life in real-time. 🚀 🔗 Project: 📄 Paper:

Humanoid robots shouldn't just follow pre-defined movements—they should perform! 🤖✨ Introducing UniAct: A unified model for multimodal motion generation and action streaming. Most humanoids are limited to pre-designed moves. UniAct changes the game by allowing robots to generate live action sequences from: 📝 Text instructions 🎶 Music rhythms 📉 Spatial trajectories 🔄 Cross-modal signals Whether it’s dancing to a beat or following a complex path, UniAct brings humanoid robots to life in real-time. 🚀 🔗 Project: 📄 Paper:

Siyuan Huang

10,654 views • 8 months ago

Introducing Jan-v1: 4B model for web search, an open-source alternative to Perplexity Pro. In our evals, Jan v1 delivers 91% SimpleQA accuracy, slightly outperforming Perplexity Pro while running fully locally. Use cases: - Web search - Deep Research Built on the new version of Qwen's Qwen3-4B-Thinking (up to 256k context length), fine-tuned for reasoning and tool use in Jan. You can run the model in Jan, llama.cpp, or vLLM. To enable search in Jan, go to Settings → Experimental Features → On, then Settings → MCP Servers → enable a search-related MCP such as Serper. Use the model: - Jan-v1-4B: - Jan-v1-4B-GGUF: Credit to the Qwen team for Qwen3 4B Thinking & Georgi Gerganov for llama.cpp.

Introducing Jan-v1: 4B model for web search, an open-source alternative to Perplexity Pro. In our evals, Jan v1 delivers 91% SimpleQA accuracy, slightly outperforming Perplexity Pro while running fully locally. Use cases: - Web search - Deep Research Built on the new version of Qwen's Qwen3-4B-Thinking (up to 256k context length), fine-tuned for reasoning and tool use in Jan. You can run the model in Jan, llama.cpp, or vLLM. To enable search in Jan, go to Settings → Experimental Features → On, then Settings → MCP Servers → enable a search-related MCP such as Serper. Use the model: - Jan-v1-4B: - Jan-v1-4B-GGUF: Credit to the Qwen team for Qwen3 4B Thinking & Georgi Gerganov for llama.cpp.

👋 Jan

691,023 views • 1 year ago

The new open-source Text to Speech model: Fish Speech 1.4 is brilliant! Trained on a massive 700K hours of multilingual speech data in 8 languages - Instant voice cloning 🗣️ - Ultra-low latency ⚡ - Compact model (~1GB weights) 🏋️‍♂️

The new open-source Text to Speech model: Fish Speech 1.4 is brilliant! Trained on a massive 700K hours of multilingual speech data in 8 languages - Instant voice cloning 🗣️ - Ultra-low latency ⚡ - Compact model (~1GB weights) 🏋️‍♂️

Rohan Paul

228,905 views • 2 years ago

🕹️ Is this the smallest language model in the world? I just managed to squeeze JAM, real artificial intelligence into 30 kilobytes, running on a 1979 Atari 800. Just A Model. Fully generative, deterministic language model, powered by a neural network and built to run on 8‑bit hardware. Atari Forever.

🕹️ Is this the smallest language model in the world? I just managed to squeeze JAM, real artificial intelligence into 30 kilobytes, running on a 1979 Atari 800. Just A Model. Fully generative, deterministic language model, powered by a neural network and built to run on 8‑bit hardware. Atari Forever.

Marek Spanel

113,782 views • 5 months ago

Microsoft just dropped VITRA-VLA, a new Vision-Language-Action model for robotics on Hugging Face. It learns dexterous manipulation from over 1 million real-life human hand activity videos.

Microsoft just dropped VITRA-VLA, a new Vision-Language-Action model for robotics on Hugging Face. It learns dexterous manipulation from over 1 million real-life human hand activity videos.

DailyPapers

19,184 views • 8 months ago

Introducing Voxtral WebGPU: Real-time speech transcription entirely in your browser. This demo runs Voxtral-Mini-4B, a powerful streaming ASR model from Mistral AI, locally on WebGPU. The model supports 13 languages and is capable of <500 ms latency. Fully private. Zero cost.

Introducing Voxtral WebGPU: Real-time speech transcription entirely in your browser. This demo runs Voxtral-Mini-4B, a powerful streaming ASR model from Mistral AI, locally on WebGPU. The model supports 13 languages and is capable of <500 ms latency. Fully private. Zero cost.

Xenova

94,356 views • 6 months ago

Excited to announce GR00T N1, the world’s first open foundation model for humanoid robots! We are on a mission to democratize Physical AI. The power of general robot brain, in the palm of your hand - with only 2B parameters, N1 learns from the most diverse physical action dataset ever compiled and punches above its weight: - Real humanoid teleoperation data. - Large-scale simulation data: we are open-sourcing 300K+ trajectories! - Neural trajectories: we apply SOTA video generation models to “hallucinate” new synthetic data that features accurate physics in pixels. Using Jensen’s words, “systematically infinite data”! - Latent actions: we develop novel algorithms to extract action tokens from in-the-wild human videos and neural generated videos. GR00T N1 is a single end-to-end neural net, from photons to actions: - Vision-Language Model (System 2) that interprets the physical world through vision and language instructions, enabling robots to reason about their environment and instructions, and plan the right actions. - Diffusion Transformer (System 1) that “renders” smooth and precise motor actions at 120 Hz, executing the latent plan made by System 2. We deploy N1 on GR1 robot, 1X Neo robot, and a large collection of simulation benchmarks. N1 achieves up to +30% boost in diverse manipulation tasks for household and industrial settings. While humanoid robots are the main focus of N1, our model also supports cross-embodiment. We finetune it to work on the $110 HuggingFace LeRobot SO100 robot arm! Open robot brain runs on open hardware. Sounds just right. Let’s solve robotics, together, one token at a time. Links to our Whitepaper, Github repo, HuggingFace model, and open dataset page in the thread: 🧵

Excited to announce GR00T N1, the world’s first open foundation model for humanoid robots! We are on a mission to democratize Physical AI. The power of general robot brain, in the palm of your hand - with only 2B parameters, N1 learns from the most diverse physical action dataset ever compiled and punches above its weight: - Real humanoid teleoperation data. - Large-scale simulation data: we are open-sourcing 300K+ trajectories! - Neural trajectories: we apply SOTA video generation models to “hallucinate” new synthetic data that features accurate physics in pixels. Using Jensen’s words, “systematically infinite data”! - Latent actions: we develop novel algorithms to extract action tokens from in-the-wild human videos and neural generated videos. GR00T N1 is a single end-to-end neural net, from photons to actions: - Vision-Language Model (System 2) that interprets the physical world through vision and language instructions, enabling robots to reason about their environment and instructions, and plan the right actions. - Diffusion Transformer (System 1) that “renders” smooth and precise motor actions at 120 Hz, executing the latent plan made by System 2. We deploy N1 on GR1 robot, 1X Neo robot, and a large collection of simulation benchmarks. N1 achieves up to +30% boost in diverse manipulation tasks for household and industrial settings. While humanoid robots are the main focus of N1, our model also supports cross-embodiment. We finetune it to work on the $110 HuggingFace LeRobot SO100 robot arm! Open robot brain runs on open hardware. Sounds just right. Let’s solve robotics, together, one token at a time. Links to our Whitepaper, Github repo, HuggingFace model, and open dataset page in the thread: 🧵

Jim Fan

467,069 views • 1 year ago

🔥 #ICRA2026 Best Paper Finalist The era of "robot VLA = single-arm gripper" is ending. Introducing Dexora — the first open-source Vision-Language-Action system for dual-arm, dual-hand, 36-DoF dexterous manipulation. 🦾 Dual Arms 🖐️ Dual Hands 🎯 36 DoF Control 🌍 Open Source Trained on: • 100K simulated trajectories • 10K real-world demonstrations Dexora achieves: ✓ 90%+ success on basic manipulation ✓ Strong dexterous manipulation performance ✓ Cross-embodiment generalization Our key hypothesis: Train on the hardest embodiment. Transfer to simpler robots later. Instead of scaling up gripper policies, we train directly in the most expressive action space and project downward to simpler embodiments. This may be a practical path toward universal robot controllers. 🎥 Demos: 📄 Paper:

🔥 #ICRA2026 Best Paper Finalist The era of "robot VLA = single-arm gripper" is ending. Introducing Dexora — the first open-source Vision-Language-Action system for dual-arm, dual-hand, 36-DoF dexterous manipulation. 🦾 Dual Arms 🖐️ Dual Hands 🎯 36 DoF Control 🌍 Open Source Trained on: • 100K simulated trajectories • 10K real-world demonstrations Dexora achieves: ✓ 90%+ success on basic manipulation ✓ Strong dexterous manipulation performance ✓ Cross-embodiment generalization Our key hypothesis: Train on the hardest embodiment. Transfer to simpler robots later. Instead of scaling up gripper policies, we train directly in the most expressive action space and project downward to simpler embodiments. This may be a practical path toward universal robot controllers. 🎥 Demos: 📄 Paper:

Hao Zhao

17,474 views • 3 months ago

Chinese robotics company Astribot released their latest World-Action Model (WAM), Lumo-2. Technical breakdown: - based on a frozen 🥶 Qwen-3.5 4B VLM - trained in 3 progressive stages: 1. Action is aligned with latent world dynamics (an abstract representation of action). Real-world actions are anchored to physical constraints, while the latent space is guided to focus on motion-relevant changes. This bidirectional relationship makes the model physically grounded -> critical for a world model. 2. Action is aligned with vision and language. Reusing the vision backbone and action encoder from the frozen VLM, the authors add a custom vocabulary (for new actions), a semantic module, an action decoder, and an action projector. This aligns the (new) action representations with the (existing) vision-language semantic space. Most importantly: it builds a direct mapping from natural-language instructions to motor execution. 3. End-to-end training on language, video, and robot data. Only the new modules (everything outside the frozen backbone) are trained end-to-end across temporal reasoning, physical understanding, long-horizon, and dexterous manipulation. At the end of the day, Lumo-2 is not the best on benchmarks, but that's not the point. What's genuinely new: - a way to combine latent world modeling and action generation through progressive alignment - a physically-grounded latent dynamics space - it lifts performance on unseen objects using un-annotated human egocentric video + Vision Pro captures, no special transfer algorithm needed Why it matters: - the whole model is thin trainable adapters (semantic module, action decoder/projector) on a frozen 4B backbone (cheap) - that scale is suited for real-time embedded inference (~2.71× decode speedup, no accuracy loss) - its real moat is long-horizon execution, where the added temporal memory pays off far more than on any other task As a result, this robot can now make your latte (5x sped up video):

Chinese robotics company Astribot released their latest World-Action Model (WAM), Lumo-2. Technical breakdown: - based on a frozen 🥶 Qwen-3.5 4B VLM - trained in 3 progressive stages: 1. Action is aligned with latent world dynamics (an abstract representation of action). Real-world actions are anchored to physical constraints, while the latent space is guided to focus on motion-relevant changes. This bidirectional relationship makes the model physically grounded -> critical for a world model. 2. Action is aligned with vision and language. Reusing the vision backbone and action encoder from the frozen VLM, the authors add a custom vocabulary (for new actions), a semantic module, an action decoder, and an action projector. This aligns the (new) action representations with the (existing) vision-language semantic space. Most importantly: it builds a direct mapping from natural-language instructions to motor execution. 3. End-to-end training on language, video, and robot data. Only the new modules (everything outside the frozen backbone) are trained end-to-end across temporal reasoning, physical understanding, long-horizon, and dexterous manipulation. At the end of the day, Lumo-2 is not the best on benchmarks, but that's not the point. What's genuinely new: - a way to combine latent world modeling and action generation through progressive alignment - a physically-grounded latent dynamics space - it lifts performance on unseen objects using un-annotated human egocentric video + Vision Pro captures, no special transfer algorithm needed Why it matters: - the whole model is thin trainable adapters (semantic module, action decoder/projector) on a frozen 4B backbone (cheap) - that scale is suited for real-time embedded inference (~2.71× decode speedup, no accuracy loss) - its real moat is long-horizon execution, where the added temporal memory pays off far more than on any other task As a result, this robot can now make your latte (5x sped up video):

Léo

32,331 views • 1 month ago

Robbyant just open-sourced LingBot-VLA 2.0 and it's much more than a bigger Vision-Language-Action model. Behind the 60,000-hour pretraining dataset run are several clever engineering ideas that push robots closer to reliable real-world deployment. Here are 6 things that stood out to me 👇

Robbyant just open-sourced LingBot-VLA 2.0 and it's much more than a bigger Vision-Language-Action model. Behind the 60,000-hour pretraining dataset run are several clever engineering ideas that push robots closer to reliable real-world deployment. Here are 6 things that stood out to me 👇

Md Riyazuddin

54,367 views • 2 months ago