A robotic hand that walks away on its own fingers! 🪬 We finally have the Thing from Addams family in robotics, and it was developed at ETH Zürich. No wheels, no legs bolted underneath, no separate drive mechanism. The same fingers that grasp things also carry the hand across a surface and hold up its weight. A quadruped has four identical legs spaced evenly around its body, and most legged-locomotion reward functions quietly assume exactly that. A hand has fingers of different lengths, arranged unevenly around a palm, with a thumb that matches none of them. The team wrote a reward formulation for that unequal geometry, and in simulation it moves faster than tuned rewards borrowed from quadruped work. They kept the hand's existing finger design and its position controller, so this isn't a bespoke machine built for the demo. And power and computation sit onboard, so it runs untethered. → Crawling, steering, and recovery after being knocked over → While holding itself up, it executes a sequence of keyboard commands with no vision at all → It pushes an object to target positions using overhead visual feedback → The simulator was calibrated against measurements taken from the physical hardware A hand that walks to a task and then performs the task is a mobile manipulator with no locomotion hardware in it. Big shoutout to authors @amrhkzp! 😮‍💨 🔗 Here's the paper: ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

A robot arm turning a valve underwater! 🌊 @ReachRobotics builds manipulators for ROVs, and the thing that makes them interesting is the actuation. Subsea arms have traditionally been hydraulic. That means an oil power pack, a lot of mass, and a work-class ROV big enough to carry all of it. Turning a single valve could mean mobilising a vessel. Reach's arms are all electric. Their Bravo weighs 4.5 kg in water, effectively neutrally buoyant, and still lifts more than 10 kg, rated to 450 metres of seawater. Control is the other half of it. A master arm with one-to-one joint matching for fine work, space-mouse command pods for driving two manipulators at once, and a 3D visualisation layer over the top. What electric actuation really unlocks is reach, in the commercial sense. Intervention used to belong to the big work-class vehicles. A capable arm at 4.5 kg moves cutting, grabbing, recovery and valve work onto a far cheaper platform. Built in Australia 🇦🇺 ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

Digit lost its bird legs! 🦿 @agilityrobotics released Digit 5 today, and the most visible change is the knees. The ostrich-style backwards legs are gone, replaced with human-like ones. Jonathan Hurst, co-founder and chief robot officer, was explicit that this wasn't a move toward looking more human: 💬 "The previous leg configuration was really optimized for very good walking and running. The new configuration is optimized for lifting heavy things and getting up and down off the ground a lot." The same logic shows up in the hands, where the argument gets I would say even sharper. He says general-purpose hands don't exist, aren't on the market, and aren't coming soon. His evidence is Cybathlon, where amputees compete at manipulation tasks using prosthetics. The winner two years running used a single-degree-of-freedom parallel-jaw gripper with one rotation at the wrist, beating competitors with five-fingered hands. So Digit 5 runs a two-finger gripper, swappable through ISO-standard mounting flanges: → Repeatable lifts up to 22.7 kg on proprietary cycloidal actuators, covering single-person lift tasks in OSHA-regulated facilities → 90 minutes of work, 9 minutes to full charge. A 10:1 run-to-charge ratio, up from 2:1 on Digit 4, which puts it past 20 working hours in a 24-hour day → 129 kg, 1.81 m tall, reaching to 2.2 m, up from 1.68 m → 360° person detection with no fixed safety zones. It slows, stops, or sits down and cuts motor power before anyone can touch it, with an independent safety controller overseeing the response First commercial availability outside North America, starting with the EU and UK, CE mark expected. Early access in H1 2027, general availability by the end of 2027. ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

. @mavenrobotics emerged from stealth with $100M to build mixed-palletizing robots for distribution centers 8 robots running 16 hours/day at 99%+ uptime. Wheeled bases at 10 mph, dual arms lifting up to 30 kg In 2024 they had nothing but "a cartoon of a robot." Won the deal against 4 competitors with working robots by visiting the customer's factories first. Palletizing alone is an $80B market!

Autonomous excavator building a wall! 🪨 This will blow your mind! 🤯 Researchers from ETH Zürich and Robotic Systems Lab have used an autonomous excavator to build a 65-meter-long, six-meter-high dry-stone wall. The autonomous system, called "Heap," precisely firstly scanned and then placed stones, forming a wall. Through computer-aided design and control, the robot was able to handle and position over 900 individual elements, some weighing over 1000 kilograms. Can we call it the Great Wall of Switzerland?! 🇨🇭 Here's the page of the project that was on from 2016 to 2025: ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

Testing the whole machining job before the first chip! 💻 Second thing that stayed with me from the @Siemens plant in Chemnitz. Offline simulation for machining is not new. What's different here is what gets virtualized. Siemens builds a digital twin of the CNC control as well as the machine, so the program runs against a virtual SINUMERIK rather than an approximation of one. That's why the workflow actually holds together. Programming, simulation and validation sit in one environment, and once a process is validated virtually, the same program goes to the real machine. → Machining programs tested virtually before anything moves → Process behaviour analysed and collisions caught before the first physical run → The validated program transfers straight into production → The MTR is controlled by SINUMERIK ONE and programmed in G-code, so the digital and physical sides speak the same language This matters more for a robot than for a conventional machine tool. Robots are being pushed into steel milling and large-part machining, where the parts are expensive, the setups are awkward, and a crash is not a cheap lesson. First-time-right stops being a nice goal and becomes the condition for using the technology at all. Robotics gives you the flexibility, the CNC gives you the control, and the digital twin is what makes the first cut the right one. ~~  ♻️ Join the weekly robotics newsletter, and never miss any news →

A humanoid on the monkey bars! 🐒 The robot from the @leggedrobotics Lab at ETH Zürich jumps up to the structure, swings across it, and drops down to a landing. Most legged robots build a terrain map first, and a heightfield is the wrong representation for a horizontal bar with air above and below it. Thin overhanging geometry is exactly what a map throws away. So the policy reads the raw scan from a head-mounted solid-state lidar, with no reconstruction step in between. An attention-based encoder picks out which of the sparse returns matter, and a GRU carries memory through the sequence, which counts when the geometry you're reaching for slides out of the sensor's view. Training is a phase-scheduled teacher-student setup. Separate privileged experts for jumping up, brachiating and jumping down, each with its own curriculum, and a scheduler handing control between them. The student is then distilled through DAgger, a critic warm-up, and PPO with the behaviour-cloning anchor decaying as reward takes over. The detail I enjoyed most sits in the sim-to-real section. Alongside LiDAR noise, they modeled battery voltage sag and actuator thermal limits. 🔗 Here's the paper page: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

🚨 Physical AI is having its "lab to production" moment, and the people driving it will be in one room later this month. If you're building or investing in embodied intelligence, Fellows Forum 2026 is where to be. On September 23–24, 500+ founders, researchers, enterprise leaders, and investors gather in Menlo Park for two days on where AI is actually heading. This year's theme: From Models to Matter. Day 2 is the Robotics + World Model track, five back-to-back sessions, all physical AI: → Foundation Models for Robotics and the Physical World (keynote) → The World Engine: Building Infrastructure for Physical AI (panel) → Scaling Laws in Foundation Models for Robotics (keynote) → Building Embodied AGI Through High-Performance Robots That Scale (keynote) → The Reality Gap: Taking Physical AI from Lab to Production (panel) The lineup is stacked: Ming-Yu Liu (NVIDIA Cosmos Lab), Dragomir Anguelov (Waymo), @peteflorence (Generalist), @Lindon_Gao & @JasonMa2020 (Dyna Robotics), @olivercameron (Oddysey AI), @startupjag (Rhoda AI) and Li Jing (Stealth), plus founders from CloudChef, Tutor, Sereact, and Walden Robotics, with NVIDIA's Ryan Julian moderating. This is the conversation the field needs right now. Robotics is moving past demos, and the hard part, getting real systems to work reliably in the real world, is exactly what these teams live every day. Fellows Fund is running this alongside @nvidia and @nebiusai. If physical AI is your world, this is one of the highest-signal rooms of the year. 🎟️ RSVP / request an invite: @AlexFellowsFund, @LucasSheiner! 🦾 ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Build a humanoid robot yourself! 🪚 OpenArm is an open-source humanoid robot.  It comes with full CAD, control code, firmware, and simulation tools, everything needed to build, modify, and operate it. The arms are designed to be compliant and backdrivable. Teleoperation is supported, with force feedback and real-time gravity compensation so operators can guide the arm naturally. Super important, on the simulation side, OpenArm works with platforms like MuJoCo and Isaac Sim, letting developers test policies in virtual environments before running on hardware.  Assemble it yourself from a kit or get it prebuilt, the goal is accessibility for research labs, small teams, and enthusiasts. The project is run by @enactic_ai in Tokyo, Japan, and aims to lower the barrier for experimenting with dexterous manipulation. Let's put robotics in mainstream! 🔥🔥 🔗 Find the link here: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Robots that climb skyscrapers to clean and inspect them! 🧼 Verobotics has built autonomous robots that clean and scan building exteriors, allowing building owners to proactively maintain and upkeep their buildings faster, cheaper, and smarter. The problem is decades old. The only way to clean or inspect a building's exterior has been to lower a person from a BMU (Building Maintenance Unit) at the top of the building. BMUs are heavy, can be out of service, complicated to operate, and limit cleaning to a couple of times per year. Their robots are lightweight, portable, and need no equipment on the roof. One person can deploy multiple robots to the building's facades and let the robots do the rest. Once they're done, the building owner not only has a clean building but also a digital twin of the building exterior for documenting health and identifying issues before they become risks. Robot spots a crack? Noted! They're equipped with onboard cameras that scan the surface, detect window frames to crawl over, and ensure not a spot is missed. That means a consistent result and thorough inspection every time. Onboard sensors continuously detect motion, location, and the surface beneath and around it. The data feeds to a hub that analyzes the surface in real time, building datasets that become the backbone of a digital twin for predictive maintenance. Love seeing the specialized robots bringing value from day 0! :) ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Six demonstrations, and voila! 🤌🏼 @BleuRobotics01 team trained a humanoid on site, the day before the doors opened at a trade show. Six human demonstrations, and that was enough. The task is a six steps with a real machine in the loop: → Pick a part out of a box → Place it into a laser engraving machine → Close the engraver → Start the cycle by pressing the two buttons → Open it again → Retrieve the finished part Roughly one-minute cycles. Exactly the kind of machine tending a lot of people do in factories today. I've visited more than 10 machine shops this year, and I keep seeing people loading and unloading machines most of the time. It then ran for four days at VivaTech at more than 80 cycles a day, at a 96% success rate. We all know that these days most of humanoid demos are usually staged. Rehearsed task, or controlled conditions. Worth noting what Bleu actually is, because it changes what the number means. They describe themselves as an AI company born on the shop floor, building universal software for deploying advanced robots into factories. It's soo refreshing to see companies who focus on deployments these days. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Construction productivity has barely moved in decades, while demand for infrastructure, energy projects, and data centers keeps growing. 🚧 During my visit to @GravisRobotics in Zurich, I spoke with CEO Ryan Luke Johns about how the company is tackling this challenge. It's worth to mention that they raised $200M Series A from SoftBank at > 1 billion USD valuation lately! 💰 Their approach is to add an autonomy layer to existing heavy equipment, particularly excavators, helping operators complete more work with the same machines. What makes this especially difficult is that construction robots don't just move through the environment, they constantly change it. Every bucket of soil alters the terrain, and ground conditions can vary from soft dirt to frozen or rocky surfaces. To address this, Gravis uses simulation to model these complex soil dynamics and train autonomous systems to dig efficiently while adapting to changing conditions. Some of the hardest autonomy problems aren't happening on roads or in warehouses, but on construction sites, where machines need to understand and interact with an environment that changes with every move. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →