Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

Can robots learn without training❓ [𝗜𝘁'𝘀 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲𝗱 ⬇ ] Teaching robots to do complex tasks WITHOUT spending hours training them. Sounds cool, right? That's exactly what DIAL-MPC does! The first training-free method for whole-body torque control using full-order dynamics: ✅ Instantly checks if a robot's moves are right... show more

Ilir Aliu

51,974 subscribers

71,502 Aufrufe • vor 1 Jahr •via X (Twitter)

Gaming Bildung Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

🚀Thrilled to share what we’ve been building at TRI over the past several months: our first Large Behavior Models (LBMs) are here! I’m proud to have been a core contributor to the multi-task policy learning and post-training efforts. At TRI, we’ve been researching how LBMs can help robots learn faster, better, and more efficiently. The key takeaways: ✅ We built an evaluation pipeline to benchmark LBM performance with real 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 ✅ Pre-training on hundreds of tasks makes models more robust—plus, we can teach new, complex tasks with 80% 𝐥𝐞𝐬𝐬 𝐝𝐚𝐭𝐚 ✅ The bigger and more diverse the pre-training, the better the results Check out our overview video, webpage and paper for more details: ✨ 🌎 📄 We hope this work helps move the field of robotics forward!

🚀Thrilled to share what we’ve been building at TRI over the past several months: our first Large Behavior Models (LBMs) are here! I’m proud to have been a core contributor to the multi-task policy learning and post-training efforts. At TRI, we’ve been researching how LBMs can help robots learn faster, better, and more efficiently. The key takeaways: ✅ We built an evaluation pipeline to benchmark LBM performance with real 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐜𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐜𝐞 ✅ Pre-training on hundreds of tasks makes models more robust—plus, we can teach new, complex tasks with 80% 𝐥𝐞𝐬𝐬 𝐝𝐚𝐭𝐚 ✅ The bigger and more diverse the pre-training, the better the results Check out our overview video, webpage and paper for more details: ✨ 🌎 📄 We hope this work helps move the field of robotics forward!

Zubair Irshad

20,377 Aufrufe • vor 1 Jahr

Robotics keeps hitting the same wall. Single task RL works, but... it does not scale to hundreds of tasks or new embodiments. This new paper looks like a real step toward fixing that. The team introduces MMBench, a benchmark with 200 tasks across many domains and robots, and Newt, a language conditioned world model trained online across all 200 tasks at once. The simple idea behind Newt: The model learns from demos to get the right priors It trains across many tasks through online interaction It uses language to ground the goal It adapts fast when a new task shows up What stood out to me: ✅ One model trained on 200 tasks at the same time ✅ Language conditioned control for both states and RGB ✅ Better data efficiency than strong baselines ✅ Strong open loop control ✅ Fast adaptation to new tasks and embodiments ✅ Full release of 200 checkpoints, 4000 demos, code, and benchmark This is a good push toward general control instead of one model per task. If you want the full paper: Project page: —- Weekly robotics and AI insights. Subscribe free:

Robotics keeps hitting the same wall. Single task RL works, but... it does not scale to hundreds of tasks or new embodiments. This new paper looks like a real step toward fixing that. The team introduces MMBench, a benchmark with 200 tasks across many domains and robots, and Newt, a language conditioned world model trained online across all 200 tasks at once. The simple idea behind Newt: The model learns from demos to get the right priors It trains across many tasks through online interaction It uses language to ground the goal It adapts fast when a new task shows up What stood out to me: ✅ One model trained on 200 tasks at the same time ✅ Language conditioned control for both states and RGB ✅ Better data efficiency than strong baselines ✅ Strong open loop control ✅ Fast adaptation to new tasks and embodiments ✅ Full release of 200 checkpoints, 4000 demos, code, and benchmark This is a good push toward general control instead of one model per task. If you want the full paper: Project page: —- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

70,090 Aufrufe • vor 8 Monaten

The next breakthrough for robot VLMs isn’t a bigger model 🧠 it’s memory. RAL accepted ✅ We show that a robot can ground a VLM using its own real-world experience, not extra training. EXPTEACH/PRAGMABOT = self-generated memory + closed-loop reflection: Fail → diagnose → replan → succeed → store experience Next time → retrieve the right experience → succeed in one shot Results on real robot tasks: 36% → 84% with reflection (STM) 22% → 80% single-trial success with long-term memory (LTM+RAG) And yes, we saw tool use emerge without explicitly asking for it. If you’re fighting brittle “VLM plans” on hardware, this is for you. #AI #Robotics ANYbotics

The next breakthrough for robot VLMs isn’t a bigger model 🧠 it’s memory. RAL accepted ✅ We show that a robot can ground a VLM using its own real-world experience, not extra training. EXPTEACH/PRAGMABOT = self-generated memory + closed-loop reflection: Fail → diagnose → replan → succeed → store experience Next time → retrieve the right experience → succeed in one shot Results on real robot tasks: 36% → 84% with reflection (STM) 22% → 80% single-trial success with long-term memory (LTM+RAG) And yes, we saw tool use emerge without explicitly asking for it. If you’re fighting brittle “VLM plans” on hardware, this is for you. #AI #Robotics ANYbotics

Haitham Bou Ammar

12,206 Aufrufe • vor 6 Monaten

30 minutes of video. Robot learns the task. Open-source, end-to-end. An open-source framework for training robot policies from only 30 minutes of human egocentric videos captured via Meta Aria glasses: Achieving zero-shot transfer to robots without any robot data collection. The method relies on Interaction-Centric Tokens that encode hand-object spatial relationships invariant to embodiment and viewpoint, supplemented by auxiliary objectives like object motion prediction and latent consistency to extract richer supervision signals from the same data. HumanEgo demonstrates strong cross-embodiment, cross-environment performance on bimanual tasks, outperforming baselines like ACT and teleop data while being trainable on a single RTX 4090 GPU. Thanks for sharing, Zhi (Leo) Wang. 📌 Website: Paper: Code: Video: ——- Weekly robotics and AI insights. Subscribe free:

30 minutes of video. Robot learns the task. Open-source, end-to-end. An open-source framework for training robot policies from only 30 minutes of human egocentric videos captured via Meta Aria glasses: Achieving zero-shot transfer to robots without any robot data collection. The method relies on Interaction-Centric Tokens that encode hand-object spatial relationships invariant to embodiment and viewpoint, supplemented by auxiliary objectives like object motion prediction and latent consistency to extract richer supervision signals from the same data. HumanEgo demonstrates strong cross-embodiment, cross-environment performance on bimanual tasks, outperforming baselines like ACT and teleop data while being trainable on a single RTX 4090 GPU. Thanks for sharing, Zhi (Leo) Wang. 📌 Website: Paper: Code: Video: ——- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

17,077 Aufrufe • vor 2 Monaten

This work makes a humanoid robot do simple parkour moves by looking with a depth camera and choosing the right move on the fly. The big deal is that it turns lots of small human moves into long, real-time robot behavior, without hand-coding every transition or retraining for each new course. A humanoid robot is usually good at steady walking, but it often fails when it has to do fast moves like jumping up, vaulting, or rolling, and then keep going to the next obstacle. The hard part is that you cannot easily collect training data for every possible obstacle shape, distance, and mistake, so robots end up learning a few moves that only work in a narrow setup. This work starts from short clips of real human parkour moves, like stepping over, vaulting, climbing, and rolling. It uses motion matching, which is basically a smart “pick the next clip that fits best right now” search, to stitch those short clips into a long, smooth plan that looks like a human doing a whole course. Then it trains a controller with reinforcement learning (RL), which means the robot learns by trial and error to copy that plan while staying balanced and not falling. After training separate expert controllers for different moves, it compresses them into 1 controller that uses only onboard depth sensing and a simple “go this fast in this direction” command. In real tests on a Unitree G1 humanoid, it can clear multiple obstacles in a row, adapt when obstacles get moved, and climb a wall up to 1.25m.

This work makes a humanoid robot do simple parkour moves by looking with a depth camera and choosing the right move on the fly. The big deal is that it turns lots of small human moves into long, real-time robot behavior, without hand-coding every transition or retraining for each new course. A humanoid robot is usually good at steady walking, but it often fails when it has to do fast moves like jumping up, vaulting, or rolling, and then keep going to the next obstacle. The hard part is that you cannot easily collect training data for every possible obstacle shape, distance, and mistake, so robots end up learning a few moves that only work in a narrow setup. This work starts from short clips of real human parkour moves, like stepping over, vaulting, climbing, and rolling. It uses motion matching, which is basically a smart “pick the next clip that fits best right now” search, to stitch those short clips into a long, smooth plan that looks like a human doing a whole course. Then it trains a controller with reinforcement learning (RL), which means the robot learns by trial and error to copy that plan while staying balanced and not falling. After training separate expert controllers for different moves, it compresses them into 1 controller that uses only onboard depth sensing and a simple “go this fast in this direction” command. In real tests on a Unitree G1 humanoid, it can clear multiple obstacles in a row, adapt when obstacles get moved, and climb a wall up to 1.25m.

Rohan Paul

37,121 Aufrufe • vor 6 Monaten

First fully ML-framework-free 3D Gaussian Splatting implementation in LichtFeld Studio. I’ve completed the migration of the full training pipeline to a custom CUDA-based tensor library. No PyTorch, no LibTorch, no autograd. Every gradient is implemented by hand, either through CUDA kernels or minimal abstractions on top. This makes it the first full training setup for 3D Gaussian Splatting with zero dependencies on existing ML frameworks. It’s not just about independence, it's about control! We now manage every byte of GPU memory, which opens the door to tighter optimization and finer performance tuning. The framework footprint is minimal, without pulling in gigabytes of ML runtime code that was never designed for real-time or graphics-driven applications. A few modules, such as the metrics and 3DGUT interfaces, are still being ported, and some operations are temporarily naïve, so performance is not yet on par with master. But this refactor lays the groundwork for: - A fully self-contained binary - Fine-grained memory optimization - Easier experimentation without the weight of an ML stack We’re getting close.

First fully ML-framework-free 3D Gaussian Splatting implementation in LichtFeld Studio. I’ve completed the migration of the full training pipeline to a custom CUDA-based tensor library. No PyTorch, no LibTorch, no autograd. Every gradient is implemented by hand, either through CUDA kernels or minimal abstractions on top. This makes it the first full training setup for 3D Gaussian Splatting with zero dependencies on existing ML frameworks. It’s not just about independence, it's about control! We now manage every byte of GPU memory, which opens the door to tighter optimization and finer performance tuning. The framework footprint is minimal, without pulling in gigabytes of ML runtime code that was never designed for real-time or graphics-driven applications. A few modules, such as the metrics and 3DGUT interfaces, are still being ported, and some operations are temporarily naïve, so performance is not yet on par with master. But this refactor lays the groundwork for: - A fully self-contained binary - Fine-grained memory optimization - Easier experimentation without the weight of an ML stack We’re getting close.

MrNeRF

50,548 Aufrufe • vor 9 Monaten

Force feedback demo Force feedback is when joystick is pushing on your hand when something is pushing on the robot arm. Feeling the force - so much helpful to control the robot, that done well it allows you to do tasks even without visual feed. You can make an experiment: close your eyes - you can easily get the headphones out of the case. Also, visual information is often not enough. For example, you're trying to pull out a usb connector, but you pull it at the wrong angle, causing it to get stuck. Visually, nothing changes, but the pressure is intense and you can break the connector. Surgical robots have been using force feedback for years, and there are also 3D styluses which use this feature, proving that the technology works and is useful. But in modern robots with AI, it's hardly ever implemented. Although it's useful for both teleoperation and AI model. That's one of the reasons why we are building our robotic arms starting with off the shelf motors rather than taking the whole off the shelf arm. There are still a range of easy wins that can be made iterating robot hardware.

Force feedback demo Force feedback is when joystick is pushing on your hand when something is pushing on the robot arm. Feeling the force - so much helpful to control the robot, that done well it allows you to do tasks even without visual feed. You can make an experiment: close your eyes - you can easily get the headphones out of the case. Also, visual information is often not enough. For example, you're trying to pull out a usb connector, but you pull it at the wrong angle, causing it to get stuck. Visually, nothing changes, but the pressure is intense and you can break the connector. Surgical robots have been using force feedback for years, and there are also 3D styluses which use this feature, proving that the technology works and is useful. But in modern robots with AI, it's hardly ever implemented. Although it's useful for both teleoperation and AI model. That's one of the reasons why we are building our robotic arms starting with off the shelf motors rather than taking the whole off the shelf arm. There are still a range of easy wins that can be made iterating robot hardware.

Igor Kulakov

18,773 Aufrufe • vor 1 Jahr

This is Marcus. Served in Afghanistan. Three deployments. Night terrors every single night for eight years. Wakes up drenched in sweat. Sometimes screaming. His VA counselor recommended a trained service dog. Wait list was four years. Cost was $25,000. Marcus couldn’t wait that long. He adopted Rocky from a rescue last October. Four-year-old German Shepherd. Rejected from police training for being “too gentle with people.” First night Rocky came home, Marcus had a nightmare around 2 AM. Rocky jumped on the bed without any training. Laid his entire body across Marcus’s chest. 92 pounds of pressure. Marcus woke up. Didn’t scream this time. Just cried quietly. Rocky stayed right there until Marcus fell back asleep. Now Rocky does this automatically every single night. Marcus showed the night vision footage to his counselor. She said: “That’s exactly the deep pressure therapy we teach service dogs. Rocky’s doing it instinctively. That’s a $25,000 behavior he was born knowing.” Rocky failed police training. But he’s saving Marcus’s life every night. Sometimes the ones who don’t make the cut are exactly who we need.

This is Marcus. Served in Afghanistan. Three deployments. Night terrors every single night for eight years. Wakes up drenched in sweat. Sometimes screaming. His VA counselor recommended a trained service dog. Wait list was four years. Cost was $25,000. Marcus couldn’t wait that long. He adopted Rocky from a rescue last October. Four-year-old German Shepherd. Rejected from police training for being “too gentle with people.” First night Rocky came home, Marcus had a nightmare around 2 AM. Rocky jumped on the bed without any training. Laid his entire body across Marcus’s chest. 92 pounds of pressure. Marcus woke up. Didn’t scream this time. Just cried quietly. Rocky stayed right there until Marcus fell back asleep. Now Rocky does this automatically every single night. Marcus showed the night vision footage to his counselor. She said: “That’s exactly the deep pressure therapy we teach service dogs. Rocky’s doing it instinctively. That’s a $25,000 behavior he was born knowing.” Rocky failed police training. But he’s saving Marcus’s life every night. Sometimes the ones who don’t make the cut are exactly who we need.

Gabriele Corno

1,020,188 Aufrufe • vor 15 Tagen

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.

The term "continual learning" has become overloaded if you see it as an ML problem. One classic thread is about memorization: regularization-based continual learning methods, such as EWC, MAS, and SI, estimate which parameters mattered for previous tasks and resist changing them too much. One modern thread is about adaptation: test-time training and inference-time learning methods, such as TTT, adapt part of the model on the incoming test stream before making predictions. These are sometimes discussed as separate threads. But in modern scalable architectures, I think they are better seen as complementary constraints: a model that learns quickly at test time also benefits from a mechanism for deciding what not to forget. In our #ECCV2026 paper, we study this in large-scale 4D reconstruction: how to build fast spatial memory that can adapt over long observation streams while reducing collapse and forgetting. Instead of using fully plastic test-time updates, we stabilize fast-weight adaptation with an elastic prior that balances adaptation and memory. Key ideas: - Elastic Test-Time Training: Fisher-weighted consolidation for fast-weight updates - EMA anchor weights that provide a moving reference for stability - Chunk-by-chunk inference for long 3D/4D observation streams We show that this scales across large 3D/4D pretraining settings, including both LRM-style and LVSM-style models, and improves reconstruction across benchmarks including Stereo4D, NVIDIA, and DL3DV-140. We release model checkpoints across different design choices: resolution, post-training curriculum, and whether the model uses an explicit 4DGS intermediate representation. - Homepage: - Paper: - Code: - Models: This work is co-led with Xueyang Yu, contributed by Haoyu Zhen Yuncong Yang, and advised by Michigan SLED Lab Chuang Gan.

Martin Ziqiao Ma

33,698 Aufrufe • vor 2 Monaten

𝗗𝗼𝗻'𝘁 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗿𝗼𝗯𝗼𝘁 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀. 𝗦𝘁𝗲𝗲𝗿 𝘁𝗵𝗲𝗺 𝘄𝗶𝘁𝗵 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀 𝗶𝗻𝘀𝘁𝗲𝗮𝗱, 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗽𝗼𝗹𝗶𝗰𝘆 Modern VLAs and world-action models can perform impressive manipulation skills, but adapting them reliably to new robots and tasks remains challenging. A natural solution is DAgger-style online imitation learning: deploy the robot, collect human corrections, and update the policy. Yet foundation models are fragile in the low-data regime, fine-tuning on a handful of interventions can improve one behavior while degrading others. Online post-training or reinforcement learning can require costly data collection and exploration, making real-world learning expensive and potentially unsafe. In our new paper, 𝗙𝗹𝗼𝘄𝗗𝗔𝗴𝗴𝗲𝗿, we take a different approach: 𝗜𝗻𝘀𝘁𝗲𝗮𝗱 𝗼𝗳 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹, 𝘄𝗲 𝗹𝗲𝗮𝗿𝗻 𝗵𝗼𝘄 𝘁𝗼 𝘀𝘁𝗲𝗲𝗿 𝗶𝘁 𝗳𝗿𝗼𝗺 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀. The key idea is 𝗮𝗰𝘁𝗶𝗼𝗻 𝗶𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻: we map human corrective actions back into the latent noise space of the frozen generative policy. These latent targets train a lightweight controller that adapts the robot while preserving the original model's capabilities. Across simulation and real robots, FlowDAgger: 📈 Learns from only 5–20 human intervention episodes 🏆 Outperforms supervised fine-tuning and latent-space reinforcement learning 🤖 Works across VLAs, diffusion policies, and world-action models ✔️ Provides reliable improvements without modifying the pretrained policy We believe this offers a practical path toward making robot foundation models improve during deployment, learning from the way humans naturally teach: through corrections. 📄 Paper: 🌐 Project: 💻 Code: This project was led by my amazing colleague Michael Murray with help from Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Harshavardhan Reddy Gajarla, Galen Mullins and Andrey Kolobov at Microsoft Research and Maya Cakmak at University of Washington

𝗗𝗼𝗻'𝘁 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗿𝗼𝗯𝗼𝘁 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀. 𝗦𝘁𝗲𝗲𝗿 𝘁𝗵𝗲𝗺 𝘄𝗶𝘁𝗵 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀 𝗶𝗻𝘀𝘁𝗲𝗮𝗱, 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗽𝗼𝗹𝗶𝗰𝘆 Modern VLAs and world-action models can perform impressive manipulation skills, but adapting them reliably to new robots and tasks remains challenging. A natural solution is DAgger-style online imitation learning: deploy the robot, collect human corrections, and update the policy. Yet foundation models are fragile in the low-data regime, fine-tuning on a handful of interventions can improve one behavior while degrading others. Online post-training or reinforcement learning can require costly data collection and exploration, making real-world learning expensive and potentially unsafe. In our new paper, 𝗙𝗹𝗼𝘄𝗗𝗔𝗴𝗴𝗲𝗿, we take a different approach: 𝗜𝗻𝘀𝘁𝗲𝗮𝗱 𝗼𝗳 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹, 𝘄𝗲 𝗹𝗲𝗮𝗿𝗻 𝗵𝗼𝘄 𝘁𝗼 𝘀𝘁𝗲𝗲𝗿 𝗶𝘁 𝗳𝗿𝗼𝗺 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀. The key idea is 𝗮𝗰𝘁𝗶𝗼𝗻 𝗶𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻: we map human corrective actions back into the latent noise space of the frozen generative policy. These latent targets train a lightweight controller that adapts the robot while preserving the original model's capabilities. Across simulation and real robots, FlowDAgger: 📈 Learns from only 5–20 human intervention episodes 🏆 Outperforms supervised fine-tuning and latent-space reinforcement learning 🤖 Works across VLAs, diffusion policies, and world-action models ✔️ Provides reliable improvements without modifying the pretrained policy We believe this offers a practical path toward making robot foundation models improve during deployment, learning from the way humans naturally teach: through corrections. 📄 Paper: 🌐 Project: 💻 Code: This project was led by my amazing colleague Michael Murray with help from Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Harshavardhan Reddy Gajarla, Galen Mullins and Andrey Kolobov at Microsoft Research and Maya Cakmak at University of Washington

Oier Mees

13,032 Aufrufe • vor 1 Monat

The future of housework just leaked on GitHub and nobody is talking about it. knox byte just open sourced a framework that coordinates swarms of Unitree G1 humanoid robots to clean your entire house on their own. It's called ARGOS. You tell it "clean the bedroom" in plain English and 2+ G1 robots split the room into zones, sweep in parallel, and sync up for the tasks that need four hands like making the bed or moving furniture. The Claude API decomposes your sentence into a task graph. An auction system makes every robot bid on every task based on distance, battery, and current load. The cheapest robot wins. Cooperative jobs go to the cheapest team. Here's what makes this different from every demo video Boston Dynamics keeps teasing: → 12 cleaning tasks baked in sweeping, mopping, wiping, vacuuming, taking out trash, making the bed, changing sheets, moving furniture, sorting items → 3 policy architectures running underneath OpenVLA-7B for language tasks, Diffusion Policy for floor coverage, ACT for dexterous bimanual work → Train it on your own footage record yourself cleaning, run one command, it extracts poses, builds a LeRobot dataset, and LoRA fine-tunes the policy → PEFA protocol for cooperative work Propose, Execute, Feedback, Adjust. If one robot fails halfway through making the bed, the team replans and retries → Full MuJoCo simulation so you test policies before pushing them to real hardware → Silver and cyan terminal dashboard that shows live fleet status, zone maps, task queues, and battery levels in real time The G1 robots talk to each other over CycloneDDS mesh using Unitree's native SDK. No cloud. No middleware. The whole thing runs on a Jetson Orin inside each robot. The wildest part is the training pipeline. Drop cleaning videos into a folder, run argos train ingest, and the framework does the entire pipeline frame extraction, pose estimation, action labeling, HDF5 dataset, fine-tune, evaluate in sim, deploy to robot. One command per stage. Unitree G1s already exist. The framework to make them clean your house just hit GitHub. 52 stars. MIT License. 100% Opensource.

The future of housework just leaked on GitHub and nobody is talking about it. knox byte just open sourced a framework that coordinates swarms of Unitree G1 humanoid robots to clean your entire house on their own. It's called ARGOS. You tell it "clean the bedroom" in plain English and 2+ G1 robots split the room into zones, sweep in parallel, and sync up for the tasks that need four hands like making the bed or moving furniture. The Claude API decomposes your sentence into a task graph. An auction system makes every robot bid on every task based on distance, battery, and current load. The cheapest robot wins. Cooperative jobs go to the cheapest team. Here's what makes this different from every demo video Boston Dynamics keeps teasing: → 12 cleaning tasks baked in sweeping, mopping, wiping, vacuuming, taking out trash, making the bed, changing sheets, moving furniture, sorting items → 3 policy architectures running underneath OpenVLA-7B for language tasks, Diffusion Policy for floor coverage, ACT for dexterous bimanual work → Train it on your own footage record yourself cleaning, run one command, it extracts poses, builds a LeRobot dataset, and LoRA fine-tunes the policy → PEFA protocol for cooperative work Propose, Execute, Feedback, Adjust. If one robot fails halfway through making the bed, the team replans and retries → Full MuJoCo simulation so you test policies before pushing them to real hardware → Silver and cyan terminal dashboard that shows live fleet status, zone maps, task queues, and battery levels in real time The G1 robots talk to each other over CycloneDDS mesh using Unitree's native SDK. No cloud. No middleware. The whole thing runs on a Jetson Orin inside each robot. The wildest part is the training pipeline. Drop cleaning videos into a folder, run argos train ingest, and the framework does the entire pipeline frame extraction, pose estimation, action labeling, HDF5 dataset, fine-tune, evaluate in sim, deploy to robot. One command per stage. Unitree G1s already exist. The framework to make them clean your house just hit GitHub. 52 stars. MIT License. 100% Opensource.

Guri Singh

27,404 Aufrufe • vor 3 Monaten

Today was my hardest workout before the Javelina 100—an uphill treadmill supercompensation session accumulating 60 minutes of intervals at 10% grade, starting at threshold and ending harder. I think uphill treadmill threshold sessions can be magical for some athletes. Threshold work is classically defined as LT2 or easier, around what you could sustain for 1 hour. In practice, that feels relatively relaxed at first, and it only starts to get harder after you accumulate a substantial amount of volume. The rationale of threshold work is that it improves lactate shuttling, helping mitochondria be more efficient at processing and transporting lactate, preventing fatigue cascades even at harder efforts on other days (or at easier efforts in marathons or ultras). In other words, it’s primarily an aerobic stress. Faster is not better. The real-world obstacles with threshold work are twofold. First, for most of us, it’s pretty slow when done right, or way too hard when done wrong. A study on the training of elite athletes found that long intervals had the lowest correlation with long-term growth, and this conundrum is probably why—athletes do their long intervals too hard, breaking themselves down without the mechanical or aerobic stimulus to justify it. Second, outdoor threshold work can be an injury risk. If I tried this workout outdoors, it would wreck my calves and high hamstrings for days. The uphill treadmill can help athletes get around these hurdles. It’s slower by design, putting the emphasis squarely on the aerobic system. That helps athletes develop a much more precise understanding of threshold. But perhaps most significantly, the uphill treadmill reduces impact forces immensely. When I finish one of these—even a supercompensation session—I feel fine the next day, allowing me to absorb way more work (and more specific work to my goals). Particularly with age, I find that running training is about managing the efforts that are high impact to be limited and focused. While most of the uphill treadmill work I do is very controlled, it’s also ok to occasionally dig deeper. Today was about supercompensation. It’s not called The Pain Cave for nothing 🔥

Today was my hardest workout before the Javelina 100—an uphill treadmill supercompensation session accumulating 60 minutes of intervals at 10% grade, starting at threshold and ending harder. I think uphill treadmill threshold sessions can be magical for some athletes. Threshold work is classically defined as LT2 or easier, around what you could sustain for 1 hour. In practice, that feels relatively relaxed at first, and it only starts to get harder after you accumulate a substantial amount of volume. The rationale of threshold work is that it improves lactate shuttling, helping mitochondria be more efficient at processing and transporting lactate, preventing fatigue cascades even at harder efforts on other days (or at easier efforts in marathons or ultras). In other words, it’s primarily an aerobic stress. Faster is not better. The real-world obstacles with threshold work are twofold. First, for most of us, it’s pretty slow when done right, or way too hard when done wrong. A study on the training of elite athletes found that long intervals had the lowest correlation with long-term growth, and this conundrum is probably why—athletes do their long intervals too hard, breaking themselves down without the mechanical or aerobic stimulus to justify it. Second, outdoor threshold work can be an injury risk. If I tried this workout outdoors, it would wreck my calves and high hamstrings for days. The uphill treadmill can help athletes get around these hurdles. It’s slower by design, putting the emphasis squarely on the aerobic system. That helps athletes develop a much more precise understanding of threshold. But perhaps most significantly, the uphill treadmill reduces impact forces immensely. When I finish one of these—even a supercompensation session—I feel fine the next day, allowing me to absorb way more work (and more specific work to my goals). Particularly with age, I find that running training is about managing the efforts that are high impact to be limited and focused. While most of the uphill treadmill work I do is very controlled, it’s also ok to occasionally dig deeper. Today was about supercompensation. It’s not called The Pain Cave for nothing 🔥

David Roche

40,681 Aufrufe • vor 1 Jahr

𝗥𝗼𝗯𝗼𝘁𝘀 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝗺𝗼𝗿𝗲 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻𝘀. 𝗧𝗵𝗲𝘆 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗳𝗿𝗼𝗺 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 — 𝗮𝗳𝘁𝗲𝗿 𝘄𝗮𝘁𝗰𝗵𝗶𝗻𝗴 𝗵𝘂𝗺𝗮𝗻𝘀. Most robot learning systems assume failure is the end of learning. In our new work, we study whether robots can improve after deployment by learning from their own failures, without any human intervention, teleoperation, or corrective labels. The key idea is simple: human videos contain structure about how the world works. We use them to learn cross-embodiment representations of action, dynamics, and value, enabling a shared predictive space between human behavior and robot experience. This allows a new learning loop: 👉 pretrain on human videos 👉 deploy robot policy 👉 observe failures 👉 reinterpret failures using human priors 👉 improve autonomously We evaluate this across 7 real-world manipulation tasks, showing: 📈 40% → 81% success rate 🏆 Strong improvements over π0.6 RECAP and RISE ✔️ Zero human intervention during post-deployment improvement 🧬 Generalizes across robot embodiments and policy backbones A key finding is that explicit failure repair significantly outperforms failure reweighting, yielding substantially larger gains under identical data conditions (+25 pts vs +5 pts on the same π0.5 base policy). Overall, the results suggest a shift in how we think about robot learning: Human videos are not only for pretraining policies. They can provide the structure needed for continual self-improvement after deployment. 📄 Paper: 🌐 Project: I am grateful for working with the fantastic leads Hanzhi Chen and Anran Zhang, and our collaborators Simon Schaefer, Kejia Chen, Shi Chen, Daniel Cremers. Special thanks to Stefan Leutenegger for co-advising this project with me. ETH Zürich TU München Microsoft Check out Hanzhi's 🧵 for more details

𝗥𝗼𝗯𝗼𝘁𝘀 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝗺𝗼𝗿𝗲 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻𝘀. 𝗧𝗵𝗲𝘆 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗳𝗿𝗼𝗺 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 — 𝗮𝗳𝘁𝗲𝗿 𝘄𝗮𝘁𝗰𝗵𝗶𝗻𝗴 𝗵𝘂𝗺𝗮𝗻𝘀. Most robot learning systems assume failure is the end of learning. In our new work, we study whether robots can improve after deployment by learning from their own failures, without any human intervention, teleoperation, or corrective labels. The key idea is simple: human videos contain structure about how the world works. We use them to learn cross-embodiment representations of action, dynamics, and value, enabling a shared predictive space between human behavior and robot experience. This allows a new learning loop: 👉 pretrain on human videos 👉 deploy robot policy 👉 observe failures 👉 reinterpret failures using human priors 👉 improve autonomously We evaluate this across 7 real-world manipulation tasks, showing: 📈 40% → 81% success rate 🏆 Strong improvements over π0.6 RECAP and RISE ✔️ Zero human intervention during post-deployment improvement 🧬 Generalizes across robot embodiments and policy backbones A key finding is that explicit failure repair significantly outperforms failure reweighting, yielding substantially larger gains under identical data conditions (+25 pts vs +5 pts on the same π0.5 base policy). Overall, the results suggest a shift in how we think about robot learning: Human videos are not only for pretraining policies. They can provide the structure needed for continual self-improvement after deployment. 📄 Paper: 🌐 Project: I am grateful for working with the fantastic leads Hanzhi Chen and Anran Zhang, and our collaborators Simon Schaefer, Kejia Chen, Shi Chen, Daniel Cremers. Special thanks to Stefan Leutenegger for co-advising this project with me. ETH Zürich TU München Microsoft Check out Hanzhi's 🧵 for more details

Oier Mees

12,379 Aufrufe • vor 1 Monat

You can't 3D reconstruct glass from images... ...WRONG! Thanks for video diffusion, now just about anything is possible! Introducing...Diffusion Knows Transparency (DKT) Transparent and reflective objects usually break robot vision and photogrammetry pipelines because they don't follow the "solid object" rules standard cameras expect. DKT is a new AI model that repurposes the "internal physics engine" found in video generation models to solve this problem. Researchers took a massive video diffusion model (WAN) and fine-tuned it using a custom-built synthetic dataset to turn it into a high-precision depth sensor. To train the AI, they built the first massive synthetic video library of transparent objects, 1.32 million frames of perfectly labeled glass and metal objects in motion. Without ever seeing a "real" labeled video of glass during training, the model (DKT) outperformed all previous specialized systems on real-world benchmarks (ClearPose, DREDS). They created a "lightweight" 1.3B parameter version that runs fast enough (0.17s per frame) to be used on actual robot hardware. Two reasons I find this project important: 1. It further proves that synthetic data will be essential for training the next generation vision models. 2. In real-world robotic tests, using DKT's depth maps nearly doubled the success rate of robot arms trying to pick up objects on tricky reflective or translucent surfaces. At home robots will need to interact with these types of objects on a daily basis. Check out the project page here: Code is LIVE! #Computervision #Robotics #AI

You can't 3D reconstruct glass from images... ...WRONG! Thanks for video diffusion, now just about anything is possible! Introducing...Diffusion Knows Transparency (DKT) Transparent and reflective objects usually break robot vision and photogrammetry pipelines because they don't follow the "solid object" rules standard cameras expect. DKT is a new AI model that repurposes the "internal physics engine" found in video generation models to solve this problem. Researchers took a massive video diffusion model (WAN) and fine-tuned it using a custom-built synthetic dataset to turn it into a high-precision depth sensor. To train the AI, they built the first massive synthetic video library of transparent objects, 1.32 million frames of perfectly labeled glass and metal objects in motion. Without ever seeing a "real" labeled video of glass during training, the model (DKT) outperformed all previous specialized systems on real-world benchmarks (ClearPose, DREDS). They created a "lightweight" 1.3B parameter version that runs fast enough (0.17s per frame) to be used on actual robot hardware. Two reasons I find this project important: 1. It further proves that synthetic data will be essential for training the next generation vision models. 2. In real-world robotic tests, using DKT's depth maps nearly doubled the success rate of robot arms trying to pick up objects on tricky reflective or translucent surfaces. At home robots will need to interact with these types of objects on a daily basis. Check out the project page here: Code is LIVE! #Computervision #Robotics #AI

Jonathan Stephens

17,712 Aufrufe • vor 7 Monaten

Robots can now reconstruct 3D scenes in real time from a single RGB camera. [📍 Projects page + paper] No depth sensor. No retraining. 30 FPS. Researchers at the Imperial College London introduced KV-Tracker, a training-free method that makes heavy models like π³ and Depth Anything 3 fast enough for real-time tracking. The idea is simple. These models use global self-attention, which is powerful but computationally expensive. KV-Tracker caches the key and value pairs from selected keyframes and reuses them for new frames. That cache becomes an implicit scene representation. Result: • Up to 30 FPS • 10 to 15x speedup • Accurate 6-DoF tracking on benchmarks like TUM RGB-D and 7-Scenes • Works with monocular RGB only It also supports object-level tracking with masks and allows saving the KV-cache for later reuse. For robotics, this reduces hardware constraints and moves real-time 3D perception closer to practical deployment. Credit to Marwan Taher (Marwan Taher) at Imperial’s Dyson Robotics Lab and many others who contributed to this! 📍 Save projects page + paper for later: Video: ——- if it matters in AI or Robotics you'll read it here first:

Robots can now reconstruct 3D scenes in real time from a single RGB camera. [📍 Projects page + paper] No depth sensor. No retraining. 30 FPS. Researchers at the Imperial College London introduced KV-Tracker, a training-free method that makes heavy models like π³ and Depth Anything 3 fast enough for real-time tracking. The idea is simple. These models use global self-attention, which is powerful but computationally expensive. KV-Tracker caches the key and value pairs from selected keyframes and reuses them for new frames. That cache becomes an implicit scene representation. Result: • Up to 30 FPS • 10 to 15x speedup • Accurate 6-DoF tracking on benchmarks like TUM RGB-D and 7-Scenes • Works with monocular RGB only It also supports object-level tracking with masks and allows saving the KV-cache for later reuse. For robotics, this reduces hardware constraints and moves real-time 3D perception closer to practical deployment. Credit to Marwan Taher (Marwan Taher) at Imperial’s Dyson Robotics Lab and many others who contributed to this! 📍 Save projects page + paper for later: Video: ——- if it matters in AI or Robotics you'll read it here first:

Ilir Aliu

53,992 Aufrufe • vor 4 Monaten

Almost 30 years ago my parents moved their young family oversees to help support pastors starting churches. They moved half way across the world and found themselves without a pastor to support. The church in this area was almost nowhere to be found. Rituals replaced relationships and the need for pastors was huge. So Mom and Dad decided to do what they said they wouldn’t do, they started a church. It was just our family at first. Felt kinda silly if I’m being honest. Just the five of us in our living room. “Why move across the world for this?” We wondered. Then, as we built relationships, the church began to grow. It was slow at first. It was also hard and lonely. But Dad was bent on building something that would train leaders and would grow without him. He always said our job was to work ourselves out of a job. The goal was to leave the leadership in the hands of the people there as quickly as possible. My parents had to move back to the States rather suddenly a few years ago after my Mom suffered a major stroke during a serious illness. The church was strong and Dad spent years discipline and training leaders. Yet, it was still sudden. We wondered, Would the church continue growing without them or begin to dwindle? Recently I decided to go back home to Northern Chile with my 3 kids to see how the church my parents started in our living room was doing. And look what I found! It’s a church of well over 500 but that’s not the impressive part. The church has planted 11 other churches, runs two children’s homes, a rehab center, a seminary training hundreds of pastors, and sends people all over the world to work in ministry. It’s been a rock in the community during earthquakes, tidal waves and civil unrest. I was able to return home and tell Dad, “Dad, your church sends their greetings”

Almost 30 years ago my parents moved their young family oversees to help support pastors starting churches. They moved half way across the world and found themselves without a pastor to support. The church in this area was almost nowhere to be found. Rituals replaced relationships and the need for pastors was huge. So Mom and Dad decided to do what they said they wouldn’t do, they started a church. It was just our family at first. Felt kinda silly if I’m being honest. Just the five of us in our living room. “Why move across the world for this?” We wondered. Then, as we built relationships, the church began to grow. It was slow at first. It was also hard and lonely. But Dad was bent on building something that would train leaders and would grow without him. He always said our job was to work ourselves out of a job. The goal was to leave the leadership in the hands of the people there as quickly as possible. My parents had to move back to the States rather suddenly a few years ago after my Mom suffered a major stroke during a serious illness. The church was strong and Dad spent years discipline and training leaders. Yet, it was still sudden. We wondered, Would the church continue growing without them or begin to dwindle? Recently I decided to go back home to Northern Chile with my 3 kids to see how the church my parents started in our living room was doing. And look what I found! It’s a church of well over 500 but that’s not the impressive part. The church has planted 11 other churches, runs two children’s homes, a rehab center, a seminary training hundreds of pastors, and sends people all over the world to work in ministry. It’s been a rock in the community during earthquakes, tidal waves and civil unrest. I was able to return home and tell Dad, “Dad, your church sends their greetings”

Bethany | Commercial Real Estate

27,587 Aufrufe • vor 1 Jahr

Model-Free Reinforcement Learning (MFRL) has been alluring, especially with supercharged compute with physics on GPU. However, the methods use 0-th order gradients, and are often not the best optimizers. Can we do better than PPO in continuous control for robotics? Turns out yes! 🥳 tl;dr: Faster, better RL than PPO in continuous control 💪 The answer lies in using more information from the simulation. We are juicing the simulation on GPU as it is, why not use it for gradients as well? This has been a driving question in a series of our works. We first studied this problem in ICLR 2022 paper on Short Horizon Actor Critic Naive gradient based methods are stuck in local minima and have exploding/vanishing gradients. SHAC solved this problem truncated rollouts and model based value estimation, where the model is Differentiable Sim. This boosted sample efficiency and wall-clock time immensely especially in high dimensional systems such as humanoids Yet, given enough compute PPO often caught up. Our follow up paper on on Adaptive Horizon Actor Critic at ICML 2024 discovers the cause and provides a fix. However, we find that even when given ground-truth dynamics, not all gradients are useful due to sample error. 1st-Order Model-Based Reinforcement Learning methods employing differentiable simulation provide gradients with reduced variance but are susceptible to bias in scenarios involving stiff dynamics, such as physical contact. We find that back-propagating through contact and long trajectories drastically reduces gradient accuracy. Using this insight, we propose AHAC to dynamically adapt its roll-out horizon to avoid differentiating through stiff contact. AHAC is a first-order model-based RL algorithm that learns high-dimensional tasks in minutes (wall clock) and outperforms PPO by 40%, even in the limit of data provided to PPO. This work is led by Ignat Georgiev alongside Krishnan Srinivasan, Jie Xu, Eric Heiden and ample assistance from warp team at NVIDIA Robotics (Miles Macklin)

Model-Free Reinforcement Learning (MFRL) has been alluring, especially with supercharged compute with physics on GPU. However, the methods use 0-th order gradients, and are often not the best optimizers. Can we do better than PPO in continuous control for robotics? Turns out yes! 🥳 tl;dr: Faster, better RL than PPO in continuous control 💪 The answer lies in using more information from the simulation. We are juicing the simulation on GPU as it is, why not use it for gradients as well? This has been a driving question in a series of our works. We first studied this problem in ICLR 2022 paper on Short Horizon Actor Critic Naive gradient based methods are stuck in local minima and have exploding/vanishing gradients. SHAC solved this problem truncated rollouts and model based value estimation, where the model is Differentiable Sim. This boosted sample efficiency and wall-clock time immensely especially in high dimensional systems such as humanoids Yet, given enough compute PPO often caught up. Our follow up paper on on Adaptive Horizon Actor Critic at ICML 2024 discovers the cause and provides a fix. However, we find that even when given ground-truth dynamics, not all gradients are useful due to sample error. 1st-Order Model-Based Reinforcement Learning methods employing differentiable simulation provide gradients with reduced variance but are susceptible to bias in scenarios involving stiff dynamics, such as physical contact. We find that back-propagating through contact and long trajectories drastically reduces gradient accuracy. Using this insight, we propose AHAC to dynamically adapt its roll-out horizon to avoid differentiating through stiff contact. AHAC is a first-order model-based RL algorithm that learns high-dimensional tasks in minutes (wall clock) and outperforms PPO by 40%, even in the limit of data provided to PPO. This work is led by Ignat Georgiev alongside Krishnan Srinivasan, Jie Xu, Eric Heiden and ample assistance from warp team at NVIDIA Robotics (Miles Macklin)

Animesh Garg

52,308 Aufrufe • vor 2 Jahren

Grip a club in your lead hand. Observe how twisting your forearm changes the angle of the face. This is largely what controls where your ball flies. Small differences in this angle lead to huge changes in shot direction. This precision is not being trained in the gym with “golf exercises”. Poor control of this face angle is what leads to a whole host of swing issues. More so than physical limitations. For example, it is far more likely you come over the top in an effort to stop the ball going right because your face is wide open and you’ve seen the ball go right the last 5000 times than it is you lack hip or shoulder mobility. It would be beneficial for me as a trainer to say otherwise, but I just don’t believe it to be true. Use the gym to work on your physical capabilities. Improve your mobility, strength, and power. These are the things you are losing as you age. Leave the swing mechanics for your practice and golf lessons. If you try to “fix your swing in the gym” you will likely do a poor job of training your swing and your body.

Grip a club in your lead hand. Observe how twisting your forearm changes the angle of the face. This is largely what controls where your ball flies. Small differences in this angle lead to huge changes in shot direction. This precision is not being trained in the gym with “golf exercises”. Poor control of this face angle is what leads to a whole host of swing issues. More so than physical limitations. For example, it is far more likely you come over the top in an effort to stop the ball going right because your face is wide open and you’ve seen the ball go right the last 5000 times than it is you lack hip or shoulder mobility. It would be beneficial for me as a trainer to say otherwise, but I just don’t believe it to be true. Use the gym to work on your physical capabilities. Improve your mobility, strength, and power. These are the things you are losing as you age. Leave the swing mechanics for your practice and golf lessons. If you try to “fix your swing in the gym” you will likely do a poor job of training your swing and your body.

Fit For Golf - Mike Carroll

227,270 Aufrufe • vor 7 Monaten