👁️The most dangerous AI failure in the DoW will not look like a hallucination. 👁️It will look formal, defensible, and audit-ready. 👁️Then it will be approved, inherited, and believed. 👁️I call that pathway the Reliability Kill Chain 👁️Why "human in the loop" is not enough:

🤖 Melania Trump Put LeRobot in School. Then the Map Got Bigger. On Sept. 15, kids at Mountain View Elementary were learning beside @NVIDIA Jetson Nano-powered robotics kits and arms that appear to be @huggingface SO-101s. Two weeks later, the surrounding strategy had moved fast. 🇮🇱 Israel’s Ministry of Education signed with @intel on AI + STEM education. 🌍 @FLOTUS Fostering the Future Together grew to 57 countries. 💰 Its tech partners committed $23M +. 🇺🇸 Federal AI language became “SI.” @realDonaldTrump @elonmusk @finkd @JensenHuang @sundarpichai @LisaSu Those elementary students reach 18 around 2033–2039. By then, the valuable robotics work may sit beyond today’s software friction: electrical repair beyond a humanoid’s dexterity, fleet recovery, safety systems, deployment architecture, and infrastructure robots cannot rebuild for themselves. @SphoenixAI 👩🏼‍💻I followed one classroom robot into the workforce these kids may inherit. ⭐️ botnews.ai/news/from-lerobot…

It Feels Through Joints 🤖 A robot hand is rotating a cube without seeing it and without tactile sensors. Researchers at ETH Zürich trained the ORCA hand to infer what is happening to the object from something every robot already has: its own joint motion. 🕙The Proprioceptive Transformer works by looking backward in time. A teacher policy is first trained in simulation with access to the cube’s actual state. That knowledge is then distilled into a Transformer that gets only a history of joint positions and velocities. No camera input. No tactile skin. No object pose at deployment. Instead, the model learns that the robot’s mechanics contain clues. When a finger loads against the cube, its measured position changes. When the cube shifts, joints deflect differently. Tendon transmission, compliance and contact alter the relationship between commanded motion and what the fingers actually do. Across time, those changes become a signal. On the real 17-DoF ORCA hand: ➡️ 11.83 rotations/min on the 55 mm cube ➡️ 3.1× the rotation speed of the proprioceptive PPO baseline ➡️ 100% rotation accuracy with zero drops in the reported evaluation ➡️ Direct joint sensing was 26.8% faster than using motor encoders The Transformer could even reconstruct cube position from proprioception alone with 13.7 mm RMSE, versus 17.9 mm for an MLP in the reconstruction experiments. There’s a useful idea here for physical AI. We usually improve robot perception by adding another way to observe the world: better cameras, tactile arrays, force sensing, depth. But physical interaction also writes information back into the machine itself. The body can become part of the observation system. That becomes especially interesting as simulation and world-model training get faster. Better models may not only let robots consume richer sensor data. They may also extract considerably more state from sensors already embedded in the hardware. The boundary is notable: this is continuous cube rotation on one tendon-driven hand, not evidence that vision or tactile sensing can be discarded across dexterous manipulation. But it gives us a much better question to ask before adding another sensor: How much of the outside world is already encoded in the robot’s own motion? Senlan Yao, Chenyu Yang, Jaehoon Kim, Aristotelis Sympetheros + @katzschmann Research @srl_ethz Article: botnews.ai/news/it-feels-… Paper: arxiv.org/abs/2605.21330

Satisfying display of every @Meta glasses ever made at #MetaConnect

Bigfoot tries the @Meta Glasses at #MetaConnect 😂

In case you can’t be at #MetaConnect here are the vibes. @muse is taking over this year

The @Muse aesthetic is immaculate @alexandr_wang 👌🏻

That’s bold. Your own folding footage repeatedly shows the system co-grasping the bedding with the garment and dragging the underlying layer during manipulation. After repeated runs the bedding shows visible distress, something no user wants. Your Skill Capture Glove has pressure/contact sensing, but layer singulation is still a publicly unresolved failure mode. You guys all have something to work toward.

@theworldlabs shipped a world model you can aim. 🎯🌍 Most video models still treat camera movement as language. You describe a crane shot, cross your fingers, and pull the lever again when the drift ruins it. 🤳Atlas, out today from World Labs, takes camera geometry as a native input type. Position, angle, path. Up to one minute of video at 1440p that holds together for the whole run. Under it: a multimodal autoregressive diffusion transformer, pretrained from scratch on text, images, video, and 3D. An LLM builds its context out of tokens in a row. Atlas builds its context out of images pinned to positions in space. Every frame it has seen knows where it stands, so whatever comes next has to agree with the geometry and not just the vibe. 📸Drop two unrelated photos into that spatial context, place them apart in 3D, and the model invents the hallway between them. ➡️ Reconstruction from one to dozens of images, no capture rig, no hundreds of dense views ➡️ Two or three views already give faithful reconstructions, and on sparse-view benchmarks the generalist beats open-source models trained only to reconstruct ➡️ Depth maps are native, so worlds come out as point clouds or Gaussian splats, the same representation already running in Marble ➡️ Third-party human raters picked Atlas over recent video models on camera following in 75 to 94 percent of head to head votes, and the advantage grows as the trajectory gets more complex Then the robotics layer: Two large environments, captured on a cell phone, 24 frames each. Atlas reconstructs the space, then generates the RGB and depth a simulated robot's body-mounted cameras would see as it moves through. The room and the robot's view of the room come from one model, so nothing gets handed off to a second system that can disagree with the reconstruction. Manipulation goes further. From a few casual recordings, Atlas helps build a scene where rigid, articulated, and deformable objects behave, then lets you swap the objects, the lighting, the positions, the motion. Scanning spaces like these traditionally required elaborate and expensive equipment. Now it is a phone in a backpack. Robots were never short on ambition. They were short on rooms to practice in. 🔗 Full technical post: lnkd.in/gJ5UPaQu 🔗 Real-to-sim background: lnkd.in/gZ_FZZwd 🔗 Early access: lnkd.in/gz8YXUMw 🔗 Marble: lnkd.in/g3KSKZ_p Congratulations to @drfeifei, @jcjohnss , @chlassner , @BenMildenhall,@KeunhongP, @ychngji6, @XRarchitect and the World Labs team, and to @YunzhuLiYZ and Changxi Zheng on the robotics side. ⭐ @BotNewsAI botnews.ai for the latest in emerging tech

You’ve heard the Microducks sing, but have you heard them sing Bohemian Rhapsody?🎶 Try the Repo: huggingface.co/spaces/FormaL… Laureen Fabre Sinègre with the help of AI used the Microduck simulator to create this masterpiece.

Human acts → robot sees → robot acts 🤖 A clean interface for robot training.🦾 @nvidia Hydra-0 uses action flow to place human hands, UMI grippers, single arms and bimanual robots in one action space: camera-plane trajectories showing where visible points move. The world model does not need their native joint or end-effector coordinates. Hydra-0 was trained on 2,202 hours of filtered multi-embodiment video. In a controlled Cosmos 2.5 comparison, replacing relative 6D actions with action flow improves PSNR, SSIM, gripper EPE, FID and FVD across all five held-out datasets; object EPE improves on all four where measured. VLM preference is mixed. Best configuration: ➡️ 90.4% lower robot-motion error ➡️ 60.2% lower object-motion error ➡️ 16× generation-only speedup after few-step distillation The interface also runs backward. A held-out HUMAN demonstration supplies desired object flow: human moves object ➡️ object flow conditions Hydra-0 ➡️ model generates compatible robot motion ➡️ supervised readout produces executable 14-DoF YAM commands No embodiment flow is supplied to the inverse model. The readout is still trained on paired real-world robot rollouts containing successes and failures. Human video is not replacing robot data 💡 It replaces task-specific expert robot demonstrations as the task specification. 🔁 RoboLab is open-loop replay. Hydra-0 starts from the first observation of an already-recorded episode and replays the trajectory the robot actually achieved. The policy is never queried on generated observations. Across 300 episodes / 5 policies / 6 tasks: - Pearson r = 0.96 - 93% per-episode agreement, κ = 0.82 These results test whether generated rollouts preserve recorded outcomes, not whether a policy can act on generated observations while model error compounds. The ~1 cm grasp miss exposes the representation’s current boundary. Limited depth awareness is one hypothesis, not an established cause. Generated rollouts can leave grasp state ambiguous: fingers may appear closed around an object without establishing that it is secured. Metric depth, tactile contact and force are not inherent to image-plane flow. Depth could recover geometry; tactile and force conditioning could disambiguate grasp acquisition, slip and force closure. The wrist-camera DROID result is qualitative. Broader camera motion, mobile manipulation and closed-loop policy evaluation remain future work. Human demonstrations specify what should happen to an object. Hydra-0 tests whether a robot can infer the motion of its own body required to produce that trajectory. @NVIDIARobotics Research: @Hongyu_Lii, @bowenwen_me, @zhu_xinghao, @YXWangBot, @du_yilun, @YunzhuLiYZ, George Konidaris, @BirchfieldStan, @SohaPouya, @80gg_overmind & @Dr_YanChang — @NVIDIARobotics with Brown, Columbia & Harvard. 🔗 lnkd.in/gmjmrcXT 📄 lnkd.in/gEyXiCSH ⭐@BotNewsAI the latest in emerging tech

Rise of the robot pets! 🐕🤖 We’re entering the cutest era of robotics. Engineers are working day and night on the physical AI problems: control, contact, sim-to-real, hardware reliability and getting learned behaviors to survive outside the lab. And somewhere along the way, we get this. @_raghuvamsi filmed one of @eyecandyrobots’ characters wandering around the Presidio, where a little girl spotted it and immediately started laughing and playing with it. There’s something joyfully familiar about that reaction. Before we learn what something is, what it costs or what it’s supposed to do, sometimes we just want to meet it. @eyecandyrobots is building physical AI characters, robots designed around entertainment, personality and interaction rather than industrial work. Founded by @_raghuvamsi, @Mankaran32 and @pr0t0_01, the team brings robotics engineers, animators and designers together to turn animated characters into physical ones. Underneath the cute exterior is serious robotics work: ☑️Animation-based reinforcement learning for character motion ☑️ Artist-centric human-robot interaction ☑️Mechanical, electronics and AI systems designed to scale across different characters ☑️ Simulation, SLAM, deep learning and control-policy deployment on real hardware The transparent little robot really looks like that. I asked Eyecandy engineer @hi_gpaul how they made its clear body: it’s 3D printed using transparent resin, followed by “a bunch of post processing.” Eyecandy is backed by @fdotinc and Lightspeed India, and the team is already thinking beyond a single character. The robots are becoming useful. I’m also very happy that some of them are becoming friends. 🦖 🔗 eyecandyrobotics.com☑️ ⭐️ @BotNewsAI stay connected to emerging tech

If you haven’t heard the Microducks sing you’re missing out 🦆🎶 @huggingface w/ @pollenrobotics just opened pre-orders for Microduck, a 25 cm, under-800 g biped that somehow manages to be both an RL robotics platform and a tiny mechanical duck I desperately want wandering around my workshop. It has 15 degrees of freedom, a camera, two IMUs, an 8×8 time-of-flight depth sensor and an articulated beak. Its policies run onboard at 50 Hz, with skills trained in MuJoCo using reinforcement learning and transferred onto the real robot. But please return your attention to the duck choir. Every Microduck gets its own generated voice, derived from its SoC serial. Put multiple ducks together and they can perform a chorale over Bluetooth. There’s no shared clock. The lowest-ID duck conducts, broadcasts the beat through BLE advertisements, and the others average their timing over roughly 25 beats to stay synchronized. They even work out who sings bass, tenor and the remaining parts without a central controller assigning them. You can also wave your hand in front of the depth sensor and turn the duck into a theremin. Obviously. The software stack is open source, including the SDK, simulator and RL training stack, so owners can retrain behaviors and deploy their own policies. The hardware itself is not open source. Pre-orders opened today at $399, with deliveries targeted before Christmas. My duck is secured 🦆 Microduck: pollen-robotics.com/microduc… Technical docs: github.com/pollen-robotics/m…

Have been building an awesome robot training data pilot. If this is of interest to you, let’s get you involved!

Need a weirdly easy side hustle? Figure will now pay approved users by the minute to do normal things inside their own homes while generating training data for humanoid robots. Taking the trash out doesn’t feel so bad when @adcock_brett is paying you #jobs #tech #news

Need a weirdly easy side hustle? Figure will now pay approved users by the minute to do normal things inside their own homes while generating training data for humanoid robots. Taking the trash out doesn't feel so bad when Brett Adcock is paying you

Two days in a row robots made us cry. The advancements happening in robotics are truly meaningful. These engineers are changing the world. Bookmark this one

#robot #nurse