{"apiVersion":"v1","generatedAt":"2026-10-25T04:22:21Z","pulse":{"slug":"robotics","name":"Robotics Pulse","description":"Robots, humanoids and automation news from X.","sections":[],"itemCount":24,"latestEditionItemCount":8,"lastEditionAt":"2026-10-25T04:00:00Z","feedUrl":"https://presto-ferry-g7bz.here.now/feeds/robotics.xml","itemsUrl":"https://presto-ferry-g7bz.here.now/api/v1/pulses/robotics.json"},"count":24,"items":[{"id":"5e99f2582a55033c","pulse":"robotics","pulseName":"Robotics Pulse","title":"HIRO Industries' Origin robot goes fully autonomous","url":"https://x.com/hiroindustries/status/2114199879585337478","synopsis":"HIRO Industries showed its Origin robot running a task fully autonomously at real speed, using a pipeline of collecting demonstrations, training a policy, and deploying it in its Station work cell. The company calls it a small task but a big milestone, and team member lexroller12 said the groundwork is laid and the focus now shifts to scale.","also":["https://x.com/lexroller12/status/2114200172817465736"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":1},{"id":"bddc675a12834190","pulse":"robotics","pulseName":"Robotics Pulse","title":"SuperMimic tracks humanoid motion with a sliver of NVIDIA SONIC's compute","url":"https://x.com/ErenChenAI/status/2114192420963279329","synopsis":"Qiayuan Liao's SuperMimic motion-tracking policy reportedly beat NVIDIA SONIC's tracking return after 3 hours on a single H100, versus about 21,000 GPU-hours for SONIC, and cut failures on a 129,514-motion benchmark from SONIC's 179 to 25, reaching zero after 9.8 hours on 2 GPUs. A separate post shows it running on the AGIBOT X2, matching SONIC's 98.4% in-distribution success with lower tracking error (19 mm vs 30 mm) after 36 hours of training on one H100.","also":["https://x.com/meetsitaram/status/2113781639189090501","https://x.com/qiayuanliao/status/2112936035911135695"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":2},{"id":"a36b297ccd91b910","pulse":"robotics","pulseName":"Robotics Pulse","title":"PaceVLA lets robot policies speed up in free space and slow down near contact","url":"https://x.com/_bbelousov/status/2114187692653293765","synopsis":"Researchers from KAIST, RLWRLD, Korea University and SNU built PaceVLA on GR00T N1.5, with a small judge that decides per action chunk when coarser, faster actions are safe, trained on VLM-written labels. On real robots it cut execution time 33–45% at similar success, and on the ALLEX humanoid it packed 6.3 instead of 3.6 packages per minute while success rose from 86% to 94%; simulation gains were smaller.","also":["https://x.com/rakybond007/status/2113580956191981982"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":3},{"id":"7c6810b1a6a59ecd","pulse":"robotics","pulseName":"Robotics Pulse","title":"Adding fingertip touch lifts a general AI model's peg-in-hole success from 31% to 99%","url":"https://x.com/Robot_AIsignals/status/2114192474264768606","synopsis":"An October 20 paper had an unmodified general AI model steer a simulated robot arm through five handling tasks with 50 expert examples each, succeeding in 469 of 500 attempts. On peg insertion, adding touch readings from both fingertips raised success from 31 to 99 out of 100, though that jump comes from a single simulated task and physical-robot tests are still to come.","also":["https://arxiv.org/abs/2610.28457","https://arxiv.org/abs/2602.10093"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":4},{"id":"977c04ac53398e67","pulse":"robotics","pulseName":"Robotics Pulse","title":"Open-source MS-Human-700 simulates a body with 700 muscle actuators","url":"https://x.com/jaskirat/status/2114193087253913639","synopsis":"MS-Human-700 is a fully open-source MuJoCo model of the human body with 700 muscle actuators, 206 joints and 90 body segments, built for muscle-level control. Researchers have used reinforcement learning with it to produce walking, running, dancing and hand movements, which makes it useful for biomechanics and humanoid control research.","also":["https://github.com/LNSGroup/MS-Human-700"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":5},{"id":"f8b9582fa1090a39","pulse":"robotics","pulseName":"Robotics Pulse","title":"Shield AI brings V-BAT, Hivemind and its counter-drone tracker to SOFEX 2026","url":"https://x.com/shieldaitech/status/2114190223676469730","synopsis":"Shield AI is exhibiting in Jordan at SOFEX 2026 with its V-BAT drone for expeditionary surveillance, Hivemind autonomy software for contested environments, ViDAR for passive wide-area search, and Tracker C-UAS, which detects air threats without radar or RF emissions.","also":[],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":6},{"id":"98b7950579a4394b","pulse":"robotics","pulseName":"Robotics Pulse","title":"New arXiv papers on humanoid VLA control and continuum-robot design","url":"https://x.com/OWW/status/2114191362379710481","synopsis":"HAF-VLA (Langzhe Gu, Chengkai Hou and co-authors) proposes a hierarchical action-flow framework for humanoid vision-language-action control. Branden Frieden, Alan Kuntz and Varun Shankar describe a probabilistic, rotation-preserving kernel surrogate for exploring continuum-robot design spaces, submitted to ICRA 2027.","also":["https://arxiv.org/abs/2610.26584","https://x.com/OWW/status/2114198070321717606","https://arxiv.org/abs/2610.26605"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":7},{"id":"45f29f84460cb8a4","pulse":"robotics","pulseName":"Robotics Pulse","title":"Rotoweld's automated spool welding setup stood out at FabTech","url":"https://x.com/JulianFried/status/2114200897756733873","synopsis":"Asked by KennethCassel what was cool at the FabTech manufacturing show, JulianFried shared video of Rotoweld's automated spool welding setup and also pointed to narrow fixture blocks from Fireball Tools.","also":["https://x.com/JulianFried/status/2114201261352648973"],"section":null,"publishedAt":"2026-10-25T04:00:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T04:00:00Z","rank":8},{"id":"986792258985b9b6","pulse":"robotics","pulseName":"Robotics Pulse","title":"LinkerBot's Linker Hand O30 holds 30 kg in a 730 g hand","url":"https://x.com/Humanoidguide/status/2114175200329613445","synopsis":"Humanoid.guide listed LinkerBot's new Linker Hand O30, a dexterous robot hand rated for a 30 kg static payload while weighing 730 g, with 20 fully active joints. It's a step up from the company's Linker Hand L6 (28 kg at 607 g), adding many more actuated joints for a modest weight increase.","also":["https://humanoid.guide/product/linker-hand-o30/"],"section":null,"publishedAt":"2026-10-25T03:00:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T03:00:00Z","rank":1},{"id":"d3a2005d9d45feba","pulse":"robotics","pulseName":"Robotics Pulse","title":"A real robot arm anyone can drive from YouTube live chat","url":"https://x.com/KuphDev/status/2114180060353810851","synopsis":"Miami Humanoids is streaming a real 7-DOF robot arm on YouTube that viewers control by typing commands into the live chat, with challenges like grabbing and stacking a cube. KuphDev tried it this hour and said the responsiveness and latency were impressive for a public teleoperation setup.","also":["https://x.com/miamihumanoids/status/2114035909750333475","https://www.youtube.com/live/QF7OO3HENzo"],"section":null,"publishedAt":"2026-10-25T03:00:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T03:00:00Z","rank":2},{"id":"5db672f918d4a948","pulse":"robotics","pulseName":"Robotics Pulse","title":"New arXiv papers on wheel-legged odometry, training-free navigation and monocular drone flight","url":"https://x.com/OWW/status/2114186167595774405","synopsis":"Hairuo Wei, Zhongyu Li and co-authors learn wheel-motion uncertainty so LiDAR-inertial odometry on quadrupedal wheel-legged robots holds up in degenerate settings where LiDAR alone struggles. ReInsVLN (Siwei Cheng, Wei Zhang and co-authors) revisits historical evidence for training-free vision-and-language navigation, and Wencan Feng, Junjie Lu and co-authors learn geometry-aware planning representations for monocular flight with cross-modal supervision.","also":["https://arxiv.org/abs/2610.26550","https://x.com/OWW/status/2114180556095209765","https://arxiv.org/abs/2610.26545","https://x.com/OWW/status/2114174325855014976","https://arxiv.org/abs/2610.26508"],"section":null,"publishedAt":"2026-10-25T03:00:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T03:00:00Z","rank":3},{"id":"213be7b45fd24574","pulse":"robotics","pulseName":"Robotics Pulse","title":"A contact-simulation demo shows a plate lifting off a rigid surface","url":"https://x.com/jvonfricken/status/2114125903638700216","synopsis":"jvonfricken followed an earlier bonded-contact video with real contact, using an augmented Lagrangian method against an analytic rigid plane, so the back of a deforming plate visibly lifts off the surface. Andres Milioto reposted it to the list this hour; accurate contact handling like this is a core ingredient for simulating robot grasping and manipulation.","also":[],"section":null,"publishedAt":"2026-10-25T03:00:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T03:00:00Z","rank":4},{"id":"d7efaa8c7bad13c7","pulse":"robotics","pulseName":"Robotics Pulse","title":"A 15-gram soft actuator that lifts 400 grams for under a toonie","url":"https://x.com/IanPritchard/status/2114164159827054930","synopsis":"Ian Pritchard showed off a new soft actuator that weighs 15 g, lifts 400 g, and costs less than a Canadian two-dollar coin to build. It was the most-liked robotics post on the list this hour, though he hasn't shared build details in the post yet.","also":[],"section":null,"publishedAt":"2026-10-25T02:00:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T02:00:00Z","rank":1},{"id":"2dab231f0e0f8e76","pulse":"robotics","pulseName":"Robotics Pulse","title":"Menlo Research's open-source Asimov humanoid, shown part by part","url":"https://x.com/clankrmedia/status/2114170058310062235","synopsis":"Clankr Media posted a laid-out photo of every component in Asimov, the open-source humanoid robot developed by Menlo Research, to show just how many parts a humanoid takes. The post is a reminder of how much hardware goes into even an open design that anyone can study and build on.","also":[],"section":null,"publishedAt":"2026-10-25T02:00:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T02:00:00Z","rank":2},{"id":"2c8355be9163dd6e","pulse":"robotics","pulseName":"Robotics Pulse","title":"Claude Opus 5.5 tries to ride a bike with a Unitree G1","url":"https://x.com/theSamPadilla/status/2114164869524345130","synopsis":"Sam Padilla shared a work-in-progress video of Anthropic's Opus 5.5 model controlling a Unitree G1 humanoid as it attempts to ride a bicycle. It's an early hobbyist experiment in using a general-purpose language model as a high-level robot controller rather than a finished result.","also":[],"section":null,"publishedAt":"2026-10-25T02:00:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T02:00:00Z","rank":3},{"id":"463a2a6f5ab0e19c","pulse":"robotics","pulseName":"Robotics Pulse","title":"New arXiv papers on bimanual imitation learning and slimming down VLA models","url":"https://x.com/OWW/status/2114168000710345044","synopsis":"\"Learn the Task, Not the Schedule\" (Shuyuan Hu, Bo Zhao and co-authors) targets spurious coupling between a robot's two arms in bimanual imitation learning, so policies learn the task instead of the timing of demonstrations. SPAR (Shenli Zheng, Li Du and co-authors) proposes structural pruning plus action recovery to shrink vision-language-action models.","also":["https://x.com/OWW/status/2114161080909848995","https://arxiv.org/abs/2610.26507","https://arxiv.org/abs/2610.26487"],"section":null,"publishedAt":"2026-10-25T02:00:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T02:00:00Z","rank":4},{"id":"329ae9b6d846c8b5","pulse":"robotics","pulseName":"Robotics Pulse","title":"An open-source ESP32 build lets an AI agent physically drive an iPhone","url":"https://x.com/davidfeldt/status/2114168355829514276","synopsis":"David Feldt pointed to a GitHub project that turns an ESP32 microcontroller into an agent-controlled keyboard-and-mouse (HID) device, letting an AI operate a phone through its normal inputs. It's a small but practical bridge between AI agents and physical hardware.","also":["https://github.com/onehumanbeing/esp32-iphone-agent"],"section":null,"publishedAt":"2026-10-25T02:00:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T02:00:00Z","rank":5},{"id":"8d626563ef59ba9d","pulse":"robotics","pulseName":"Robotics Pulse","title":"Robot gripper logs beat vision critics at judging robot success","url":"https://x.com/antopatrex1/status/2114144530652094476","synopsis":"Anto Patrex of CosmicBrain AI shows that a robot's own gripper log, the gap between commanded and actual finger closure, can tell whether a grasp, lift, carry and release really happened, with no camera, model or labels. On 1,000 human-judged RoboArena A/B sessions the log picked the preferred rollout 73% of the time, versus 70% for a frontier vision model and 56% for NVIDIA's Cosmos Reason 2 (8B), and adding the log cut Cosmos's error by 41%. The write-up also reports pre-registered misses, including that curating \"clean\" demos made policies worse (11% vs 25% success).","also":["https://x.com/Cosmicbrainai/status/2114145562555777370"],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":1},{"id":"027bd1a8a29f8a01","pulse":"robotics","pulseName":"Robotics Pulse","title":"Berkeley's SuperMimic trains a Unitree G1 humanoid on one GPU in three hours","url":"https://x.com/Robot_AIsignals/status/2114143652473606528","synopsis":"An Oct 20 UC Berkeley paper says SuperMimic trained motion-tracking software for a Unitree G1 in three hours on a single H100 and beat the earlier SONIC system, which used 128 chips for seven days, by rehearsing movements inside a learned model of the robot's body instead of a full physics simulator. On a real G1 it performed a side roll and a 360-degree flip, though the post notes the real-robot evidence is only four moves on flat ground with no published success rate.","also":[],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":2},{"id":"9144362312f9b774","pulse":"robotics","pulseName":"Robotics Pulse","title":"Andrew Kang publishes a deep dive on the \"ChatGPT moment\" for robotics","url":"https://x.com/Rewkang/status/2114149925307269465","synopsis":"Andrew Kang released a RoboStrategy report on how capable physical AI models are today and what the robotics adoption curve looks like, based on conversations with most of the major players. It was the most-engaged robotics post on the list this hour.","also":["https://robostrategy.co/news/the-chatgpt-moment-for-robotics"],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":3},{"id":"ca535ce7ecd3bf72","pulse":"robotics","pulseName":"Robotics Pulse","title":"Microduck demand is causing a Dynamixel servo shortage for hobbyists","url":"https://x.com/superintel_xyz/status/2114146737724219580","synopsis":"Robot builder @_Stocko_ says his new 17-motor PhoneGirl robot faces a two-month wait for Dynamixel XL330 servos because Pollen Robotics bought up the supply for Microduck. At 15,000+ units sold and 15 motors each, superintel robots estimates Microduck has absorbed about 225,000 of those servos.","also":["https://x.com/_Stocko_/status/2114084333975699804"],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":4},{"id":"0deb16fdcfa37a98","pulse":"robotics","pulseName":"Robotics Pulse","title":"Why failure data, not more clean teleop demos, may be robotics' real bottleneck","url":"https://x.com/rahim_mirani/status/2114157513344155722","synopsis":"Abdul Rahim Mirani argues teleoperation mostly captures the easy 90% of a task because operators simply retry failures without logging them. He points to Physical Intelligence's pi*0.6 espresso results, where adding about 400 human corrections plus the robot's own failed runs more than doubled throughput and pushed success past 90% on the same model and robot.","also":[],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":5},{"id":"a902205b09bbed9f","pulse":"robotics","pulseName":"Robotics Pulse","title":"A ranking of 30 Chinese humanoid robot leaders","url":"https://x.com/techniahqrobot/status/2114159069523034413","synopsis":"Techniahqrobot posted a ranked list of 30 Chinese humanoid robotics companies led by AgiBot, Unitree, UBTECH, Galbot and Leju, with XPeng, Xiaomi, GAC, Chery's AiMOGA and Tencent Robotics X also included. The author says a second graphic and thread are coming to complete the list.","also":[],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":6},{"id":"3b31b9b70ad35372","pulse":"robotics","pulseName":"Robotics Pulse","title":"New robotics papers: ALIGN and an EMG-driven elbow exoskeleton study","url":"https://x.com/OWW/status/2114155836314370100","synopsis":"ALIGN (Zechen Wang and colleagues) aligns how conditions are estimated and expressed in advantage-conditioned behavior cloning for robot learning. A separate human-in-the-loop study from Francesco Pascucci, Andrea Calanca and co-authors compares proportional control with adaptive gravity compensation in EMG-driven elbow exoskeletons.","also":["https://x.com/OWW/status/2114149210182263161"],"section":null,"publishedAt":"2026-10-25T01:00:00Z","firstSeenAt":null,"editionAt":"2026-10-25T01:00:00Z","rank":7}]}