Launch HN: Vassar Robotics (YC X25) – $219 robot arm that learns new skills
578 points by charleszyong 7 days ago
578 points by charleszyong 7 days ago
Hi HN — I’m Charles from Vassar Robotics (https://vassarrobotics.com/ - not much there but you can order the robot at https://shop.vassarrobotics.com/products/navrim-robot-that-l...)
Edit: the entire run sold out thanks to HN today—thank you all! And sorry to anyone who missed out. You can get in on the next batch here: https://vassarrobotics.com/newsletter.
We are bringing an upgraded version of the long beloved SO-101 robot arms to a $219 price point with improved mechanical design and added intelligence. See what it can do here: https://youtube.com/shorts/xNyPKJZI400 (demos are sped up as shown in the video)
I’ve spent a few years building RC planes (https://cyo.ng/hangar/) and micro gas turbines (https://set.mit.edu), and I’ve always wished hardware were cheaper so more people could experiment.
I’m now launching a $219 desktop robot-arm kit that keeps LeRobot SO-101’s kinematics, swaps key parts for sturdier, more precise SLA prints, and adds two integrated 480 p cameras. After plenty of supplier haggling, the whole kit costs less than the twelve servos alone. I’ll release the updated mechanical design under an MIT license by June 30.
On the software side, I'll also release an MIT-licensed MCP server by June 30 that exposes the local robot policy as tools for agentic LLMs (Opus 4, o3, etc.) to use in long-horizon tasks. Here's how it works: You can teach the robot new skills through teleoperation. During inference, you simply talk to the agentic LLM using natural language instructions. The LLM then calls the local robot policy through MCP, automatically decomposing your high-level requests into executable robot commands.
Thanks to the LeRobot community for making such an amazing robot accessible. If you’ve contributed to the LeRobot GitHub repo, email hello@vassarrobotics.com for a 20% discount coupon as a small thank-you.
I’d love your feedback! Beyond manufacturing, cleaning up the codebase, and writing docs, I’m considering: a force-controlled gripper, a parallel-jaw gripper, an extra wrist DOF (matching the new Trossen and ARX arms), full force feedback on the leader arm (though that may triple the price), a more affordable version with lower resolution each joint, and a longer-reach variant. Which of these—or something else—would be most useful to you?
You can order it here if you want: https://shop.vassarrobotics.com/products/navrim-robot-that-l.... (Edit: sold out! You can get in on the next batch here: https://vassarrobotics.com/newsletter. I hope we can have your business in the future.)
Looking forward to any and all comments!
---
Edit: A quick explanation regarding shipping times (as stated on our shop page):
• The first batch of 20 units, which will be shipped by June 30, is sold out.
• The second batch of 100 units will be shipped by July 15 (unassembled kits) and July 21 (assembled units). The order limit is to ensure we can ship on time and maintain high quality.
For those who have already placed orders: I will reach out individually to ask if you would like to receive weekly progress updates from now until the shipping date.
Sorry for being MIA for a bit. All 120 units are now sold out. I’ve created a waitlist at https://vassarrobotics.com/newsletter to keep you updated on when the next batch will be available (most likely in late July). Thank you so much for everyone’s support. My top priority now is to get all the orders shipped on time and with high quality. Looks like you might've found that magical thing which is product-market fit. I'm rooting for you. $200 is a really nice entry-price point. If I'm being honest, I'm marginally interested in this, but doubt that I'd actually use it too much. But the price point is justifiable for someone who is just interested in it from a learning point of view (if I bought this, used it a few times and learned a bit more about AI as it relates to robotics, it'd have paid itself off easily). Rooting for you to succeed. Thank you for your support! I’m actively working on the design and building the supply chain. In the next 3 to 6 months, I should be able to: 1. Keep the price similar while adding new features, such as a torque-controlled gripper.
2. Re-examine design decisions to see if we can offer a similar product at an even better price. Firstly, at the $219 price point you can have my money already. Beyond that, things that appeal to me are basically anything which increase the likelihood I can accomplish high dexterous fine motor control skills, for things like tinkering and DIY assembly. I think that would include extra wrist DOF and a longer-reach variant. Integrated cameras are an interesting idea, but I'd like to be able to swap them out for my own. My dream is to have some sort of multi-arm table at home. I imagine holding a circuit board, small component, soldering iron, and wire with four robotic arms I control with shaky hands from my laptop. :D So true. Every time I solder surface mount components, I always wish I could have a steady hand. Sadly, this arm doesn't have that kind of accuracy. The output shaft of the servos we use has about 1 degree of wiggle room and the mechanical structure adds more. To get better accuracy, if sticking with this kind of RC servo, it's basically required to have two servos per joint to preload each other to kill that wiggle room. It's something I've been calculating, but I just can't figure out a way to offer it at a good price. Interestingly, for arms that are popular in academia, even when the price goes to $10k (like ARX or Trossen), the wiggle room is still there (better, but still there). I was recently trying to get better angular accuracy with servos and minimize backlash. One option that kind of worked was to have a pulley on the servo shaft which wound a string attached to a spring to add mechanical bias. But I ended up giving up and going with 400 step stepper motors instead. They're larger, draw more current, and the drive circuitry is more complicated (it can't get simpler than a PWM servo after all). But they're accurate and significantly quieter. Here's how to do it: https://www.youtube.com/watch?v=GCHXNcpq3OA Check his channel for servo motor modifications. Oh, I think I actually saw this! But it's a very manual process with a lot of small parts to assemble, and I want to eventually design a product I can manufacture myself so I didn't pursue it. I understand. But you could sell a normal version and an enhanced version for those who need the accuracy. I know nothing about the topic but could one build some kind of 3d pantograph to scale down the motions? What you need is to scale down the tolerances. To remove the wiggle room. One of the solutions that does not add a bias that I remember is two identical flat gears on the same axis with a spring that tries to rotate them one relative to another. This removes the wiggle room between this composite gear and the next, regular gear. The motor may have wiggle room, but the gears (which carry angle sensors, don't they?) move without wiggling, and react immediately as you reverse the direction. The load is limited though: the beating surface is twice as small, and the friction is higher. Here is that animation you've never asked for. https://img.go-here.nl/unwiggle.gif One could put those in series too and get even less range of motion in exchange for less wiggling. One could also duplicate the contraption on both sides. Then could replace the arm with cables (under tension) and control motion further down the arm. Furthermore it seems you could remove the motors from the moving parts? Ill let myself out. Motors inside the big body and limbs actuated by series of strings is like many arthropods are powered. Evolution approves! You could also control the tension of the wires and have accuracy on demand without the wear. I have this mental image of a darts player moving his arm back and forwards to smooth out the tension. I sometimes wonder if it makes more sense to just use those yellow gearboxes, everyone seem to start with SG90s only to reimplement most of the servo part. have to share the noob thought because it is funny: You could attach an unbalanced wheel to a motor and induce a vibration to maximize wiggling and frequency across the available w-room. Backlash is a real problem. I’m working with the supplier to prototype a small batch with 20 to 30% less backlash, though this may come at a slightly higher price and potentially a shorter lifespan. If cost isn’t a concern, harmonic drives combined with brushless servos are excellent. I have a few harmonic drive units, and they’re truly amazing. There are cheaper options for soldering SMD-parts on prototype boards. Developing and teaching robot arms to do it would give a good demonstrator but economically it’s a disaster. And mass production is already highly automated. > it's basically required to have two servos per joint to preload each other to kill that wiggle room Couldn’t you preload with some form of spring? The advantage of servos is that they can provide constant torque preloading, allowing the preload to be kept low (otherwise, the servo will overheat) but still sufficient. A spring might also be an option if designed properly. I’ll probably give it a try in July. You can use a spring with long travel and low K (like a very soft torsional spring, preloaded several turns, or a spring connected by a wire to a drum on the joint) to reduce the change in torque across the range, if that helps? even something twice the price ($438) would still be a great deal. Mind telling us something about your pricing strategy trade-off consideration matrix? Design for manufacturing is one thing. I did it a lot when building micro gas turbines in college. Sometimes changing the design or manufacturing process will make it 10x faster to make one while not compromising the performance. The second thing is low margin. When people are pricing hardware, they usually plan a 50% to 100% margin to offset various costs that happen in the real world. From what I've heard, in extreme cases, some products cost around $100 while they are being sold at close to $1000. I believe in the Prusa printer approach: you design a good product and price it a little bit above cost. So the company grows with the community. Deep down, there are so many times that I wish I could afford a fancy tool like a Milwaukee drill or a Mitutoyo caliper. And in extreme cases, I really wish I could have a HAAS UMC-400 or even a KERN Micro HD+. Now that I can set the price, I really wish I could make someone get what they want without breaking their bank. For folks curious about why these kinds of margins are built in for hardware products, I found these videos informative: https://www.youtube.com/watch?v=UwrkfHadeQQ As a customer, thin margins scare the crap out of me. It means you probably won’t be around in a few years. Seems like this stuff is mostly built on open source, so no need to be too scared. That’s not at all reassuring to someone who wants the benefit of the product, not the obligation to become an expert on the product. I’ll buy one of these for sure, but I would cheerfully spend 3x the price if it meant being sure of support and repair and software updates for a few years. I have bins full of exciting devices that no longer integrate, many of which have open source communities, which I just don’t have the time to deal with. This is based on the open-source HuggingFace/LeRobot SO-101, so it'll probably be possible to fall back to that. Or you can just get an SO-101. But I think this is attempting to draft on that ecosystem, and this is cheaper than Alibaba parts to build an SO-101. that's more scary, not less! Sure, if the worst happens, since it's open source, we'll have the source, but being open source means they've got to figure out a more complicated business model than make thing, sell thing, profit. there are some success stories but also a lot of failures. I guess I should clarify that it's open source, and seems to be essentially forked from HuggingFace's SO-101, so it's probably not too hard to fall back to that stuff (with fewer features). It’s worth reading the history behind Raspberry Pi. It depends on the team. With the right product market for, this strategy has and will work. In my view, the Vassar team got the pricing strategy right to do something even bigger than Raspberry Pi if they can scale at this price point for the low end model. For software support, the software is open source, so the community can keep it going even if we’re no longer around (hopefully not anytime soon). For repairs, the components are inexpensive enough to simply swap in a new one, and you can always find replacements on AliExpress. Well, based on that info, I guess usability for pick-&-place of surface-mount components on PCBs is out of the question. Yeah… the accuracy isn’t good enough. A gantry system is probably needed. There was one robot startup (Haddington Dynamics) figured it out how to do it at a higher price. Sadly they've been acquired and shifted directions I think. I had that same thought. I travel a bunch for work and would love to be able to wire up a RPi/Arduino remotely to a sensor or other device and run a test or two. You need some technical specs on the website. How many DOF does it have? Does it have joint angle sensing? If so, what's the resolution? What's the interface to the servos? What's the payload capacity? Does it have integrated motor controllers? How long is it, and what does the dexterous workspace look like? As a roboticist, what I'd vote for, in order, is: - more degrees of freedom - interchangeable tools, either an actual tool changer (unlikely at the price point) or a fixed bolt pattern with electronic passthroughs - better joint sensing, e.g. absolute encoders, joint torque sensing - fingertip force sensing Thank you for the feedback! Thinking out loud:
• Adding one DOF to match ARX kinematics is doable, with a price increase of $30–40. • A tool changer is a great suggestion. A few of my friends are working on kinematic couplings, which would be ideal for this. I’ll need to give some thought to how to pass electrical signals and power to the tool, while also keeping it lightweight. • Could you share what functionality you want in terms of encoders? The ST3215 uses 12-bit magnetic encoders, which can retain position after power loss. Are you looking for higher resolution? For torque sensing, if the order volume is large, I can add this for just a $20-30 price increase. • Finger tip force sensing: Is this for applications like picking up an egg? Just to clarify, these improvements is for future models. You'd also ideally add torque sensing at the joints because it opens up a whole world of control techniques that you can't get with just joint position sensing. You can do compliance or force control, which lets the arm act as if it had a spring at the joints, so when it hits something the impact is nice and gentle, and importantly, so you can do things like e.g. a bolt insertion task where you have to control the position of the arm in x and y but you want to exert a small positive insertion force in z. Or if you're folding clothes or handling cables or wires or anything else flexible, you really need to have a sense of touch. You can't really do these tasks very well with position sensing and vision alone. Another idea: Maybe add a passive mounting adapter and power leads at the end effector so people can add their own vision or lidar sensors, and just let them connect via bluetooth, so you don't have to route signal cables? FYI, I am a space roboticist by trade and I teach a graduate level class in robotics at the University of Maryland. Also, for the type of work I'd do with an arm like this, I'd be more than happy to just have the follower arm. You need a leader arm to do some types of teleoperation or imitation learning, but not really to do reinforcement learning or learn about control theory. What you do need is an articulated rigid body model that you can import into e.g. NVIDIA Isaac Lab or Gazebo. The availability of a good digital model is a HUGE selling point. https://github.com/TheRobotStudio/SO-ARM100/tree/main/Simula... I hope applies to this first gen of the product. > Maybe a modular design ? Pardon my naïve imagination but, would stackable joints work - with same connector as the extremity tooling ? The joint would be a standard piece and more degrees of freedom would just mean stacking additional joints. I suppose this has already thought about... Check out Mill-Max magnetic connectors for the tool connectivity. Very good suggestion! They look really promising. I’ll talk to the company to see if the price will be acceptable. > You need some technical specs on the website. How many DOF does it have? Does it have joint angle sensing? If so, what's the resolution? What's the interface to the servos? What's the payload capacity? Does it have integrated motor controllers? How long is it, and what does the dexterous workspace look like? The post says "kit that keeps LeRobot SO-101’s kinematics" so it's probably very similar to [1] namely 5DOF and a gripper, using STS3215 servos [2] > As a roboticist, what I'd vote for, in order, is: As they are making a robot at the $219 price point, I very much doubt they have the money to add anything to the design. [1] https://huggingface.co/docs/lerobot/so101
[2] https://uk.robotshop.com/products/magnetic-encoding-servo-st... Thank you for stepping in. Yes, it’s 5 DOF and a gripper using ST3215 (12V for the follower arms and 7.4V—various gear ratios—for the leader arms). As for hardware features, we can’t add much to the current model since, as you mentioned, we are running on very thin margins. We’re gathering suggestions primarily for future models. They did ask for suggested upgrades. “I’d love your feedback! Beyond manufacturing, cleaning up the codebase, and writing docs, I’m considering: a force-controlled gripper, a parallel-jaw gripper, an extra wrist DOF (matching the new Trossen and ARX arms), full force feedback on the leader arm (though that may triple the price), a more affordable version with lower resolution each joint, and a longer-reach variant. Which of these—or something else—would be most useful to you?” Are there any affordable robot kits you recommend for learning control, CV, RL etc.? I was budgeting for the SO-101 so I think I'll get OP's device and then something that's not an arm for variety. Love this !! I have been searching for a homegrown store selling the so101 and other open source robots. Took me 6 weeks to get my unassembled kit for ~$250 from wowrobo (and it got stuck in inspections at the border). Would be cool to connect to learn more about your plans and offer some suggestions for improvements based on my experience so far. Of course this arrives right after I order all the electronic parts and just kicked off the 24+ hour 3D print job to complete my SO-Arm101. But I’m routing for you! Thank you! Let me know how you like the SO-101 design. If you have complaints, I might be able to find a way to fix it ;) Literally same. Just finished printing the leader arm and not I have another 20 hour print for the follower. Curious where you sourced the parts? In Canada, shipping kills it for me. When I priced out the robot + electronics + $100 in shipping, I am around $700 - far cry from the $100 on the "sticker". You should put it on Amazon; we used a robotic arm in one of the classes I taught, and for logistics reasons it was basically the only way we could order stuff. Plus it helps with discovery. I'm sure there's an extra fee but it's sometimes just impossible to order things if you're a big organization from small sites like this. Thank you for the suggestions. I hear you. When I was at college, the school system basically only allowed Amazon plus a few industry-specific suppliers.
Please allow me to prioritize manufacturing and testing so that I can ship the product as soon as possible and with the highest quality possible. Then I will start expanding sales. Also, the servos we are using actually have a version that has lower torque/force output, which would be safer for students but also limit what they can do with it. Would you be interested in the "safer" version for classes? I was just a TA, but the big issues for my professor were cost (a lot of arms cost 1000+, so if you’re buying 30 then that’s $30k+ which is hard to budget for) or it was number of axis. Since it was a class he wanted more motors, which made the arm more complex and more similar to the real world. Counter-intuitively to teach someone about robotics you want an arm that’s complex so students have more to learn, students would have to implement their own kinematics and some of the affordable ones were too simple. I never heard him mention anything about motor safety, and when we had to fix them we would often buy stronger, metal gear servos as replacements. I love the product though, and I appreciate your input! Can you explain more how this is possible? For a layman like me, what is happening when you tell the robot to do something and how does it know it's going to the right place? SO-ARM101 has a leader-arm, which is the arm with same exact dimensions and same servos - but used to read/record the trajectory. You move it with your own hand and teleoperate the follower-arm in real-time. Follower-arm is visible in the demo videos. If you fully control the environment: exact positions of arm-base and all objects which it interacts with - you can just replay the trajectory on the follower-arm. No ML necessary. You can use LLM to decide which trajectories to replay and in which order based on long-horizon instruction. Yes. You are exactly right. If you want the model to have some adaptability, you will need to train a policy like ACT or GR00T. Just a quick difference I need to point out as it's critical product spec: leader arms are using 7.4v version of ST3215 (various gear ratios) while follower arms are using 12v version of ST3215. (12v version have higher peak torque at close to 3 Nm) > various gear ratios for anyone following along at home: THIS IS NOT A SMALL DETAIL! there are 12 servos total and 4 different types of 7.4 volt ones in the box. make sure you use the right one, or else you'll waste precious time to reassemble the arm. Thank you for confirming! Love how simple yet magical your demos look, the elegance of bridging LLM-driven long-horizon planning with the arm. Wait so the arm isn't doing any learning or moving on its own? I don't understand why you need a leader arm? It sounds like you use the leader arm to show the robot how the task should be done. If you just used your own arm for the task the robot would have to translate human movements to its own mechanics (hard) but this way it does only need to replicate the movement you showed (easier). After you teach it how to do the movement it can then do it by itself. You show once and it can repeat a million times. Ok I was under the impression (due to the cameras) that it's doing something with machine learning or can do a novel movement. This is just recording movements and playing them back. If you bridge recorded trajectories with LVLM, then cameras are necessary visual input for LLM to decide which sub-tasks need to be performed to accomplish long-horizon task, and sub-tasks correspond to pre-recorded ("blind") trajectories which are replayed. If you go beyond pre-recorded "blind" trajectories into more robust task-policies (which you would have to train from many demonstrations) then cameras become necessary to execute the sub-task. It's just bunch of motors on a stick. Doesn't come with a computer at all. But that's still worth >$200, as 1) building an arm that works is a project of its own, and 2) hardware standardization is crucial for code reusability. ...then why is this titled "that learns new skills"? It's intended as a close replica of HuggingFace SO-101 arm, so same models should work. It also comes with a replica leader arm for "teaching", or teleop based programming in layman's term. Wow! Recently my son has been asking about doing a project with a robotic arm, and this looks amazing, especially at the hobbyist-friendly price point. And adding in AI is really cool - and just the thing to really grab the attention of an eight year old boy :) Will these be available in the UK, perchance? A bit of an aside, but how hard is it to get into building RC aeroplanes, compared to FPV copter drones? RC aeroplanes need some practice and a bigger field compared to FPV drones. I think I spent a week flying in simulators and another 2 weeks crashing several times to get a basic hold on it. It's kind of like training a robot foundation model to learn a new embodiment That being said, I enjoyed every moment flying my planes. I built and flew quite a few quadcopters but they never felt that free because there's always that control algorithm between the pilot and the motors, while aeroplanes are basically just mapping the movement of the joystick to the servos. I believe the UK has a lot of great local clubs, and I believe that's the best place to get started. Side note, when your son gets more experience in the field, he might wanna build his own gas turbine to power his planes. And this association based in UK is the best on this planet: https://www.gtba.co.uk For UK delivery, let me look into how to set up international shipping. Will get back to you by end of the day. I hadn't thought about clubs, probably because I live in a small, rural Scottish town... but I just had a quick look, and incredibly there's an active club just a few miles from me, which I had no idea even existed! If the goal is the building. Balsa kits (an xacto knife, 2 bottles of super glue [thick/thin], CA-accelorator) are the way to go. Discuss gliders are easy to manage the risk of learning how to fly, and are light, so crashes will only be mildly catastrophic. I have this one, and it was easy-ish to build (~20 hours?) http://wrightbrothersrc.com/products/gambler.htm If the goal is the flying. You can't go wrong with an easy star. I've crashed mine a million times. You just patch it back together humpty dumpty style with thick CA + accelerant. Bonus points for the prop being in the back, so if you run into stuff you (probably) won't draw blood. https://mrmpxhobbies.com/product/rr-easystar-3/ Note that the hobby does require some skill w/ flying and need some level of risk management. There are cords that let you plug your transmitter into a computer/fly over a simulation that can help with the former. > 2 bottles of super glue [thick/thin], CA-accelorator) I haven't built a balsa wood plane in ages. But so, the glue of choice has changed ? No more balsa wood glue with atrocious fumes ? Building RC planes is a little harder IMO, but not much. The main difference in building planes is you have to pay attention to center-of-gravity much more; minute differences will make the difference between your plane flying amazingly, like a brick (nose heavy), or not at all (tail heavy). There's also more work to do in setting control linkages and surface throws. But, overall, it's not too tough with most models. Takeoff with planes can be very stressful the first few times; you have to choose between ground/runway takeoff, which typically results in a very inefficient model due to landing gear drag and is prone to flipping over, throwing the plane by hand, which requires practice and can be quite hazardous with a "pusher" style plane with the prop at the back, and building some kind of bungee launcher, which you then have to set up and lug around. Then you have to decide how to fly - line of sight or FPV. Line of sight flying is quite an acquired skill and has a very steep learning curve - you basically have to learn to "become the plane" and understand how your control stick inputs are affecting the attitude of the plane without being able to see it very well. FPV plane flying, while less popular than LOS, is very easy and much more rewarding IMO. The reaction time in all but the most extreme plane stunt flying is much less dramatic than in FPV quads. And, due to quirks of the general hobby flight control software scene, most hobby FPV planes have a working loiter-in-a-circle setting while most FPV quads have a barely-functional GPS rescue mode and little to no ability to actually hover (it's very rare for an FPV quad to "just stay put"; this is the realm of camera drones). I fly FPV quads when I need a focus/adrenalin boost and FPV planes when I just want to relax and chill. You can fly planes in an adrenalin style, but they're much more conducive to just looking at the scenery and goofing around. Massive bonus points that most plane builds are almost silent compared to an FPV quad so you don't worry about bothering people so much. Planes, like quadcopters, are as complicated or simple as you want them to be. They're available fully ready to fly, as kits with different levels of work needed, or you can build from scratch and choose your own parts and design. Flying is pretty different, though. If you're used to a copter that will just stay put when you release the controls, flying planes will be an adjustment. Yes yes! Flying an aeroplane has no pause button. You are on your own from taking off to landing. It's a great practice not to panic under stress (I never flew one but I guess racing FPV quadcopters probably has the same feeling) I’ve just set up shipping service to the UK. You should be able to place an order now. Let me know if you have any questions! While I don't think this will ship in time. There is a global online hackathon using these robot arms on Hugging Face June 2025, 14-15. https://huggingface.co/LeRobot-worldwide-hackathon Man, I gotta say that I tried really hard to see if I could ship before that, but I failed ;( Chasing suppliers is pretty much like dating: sometimes no matter how hard you try, you just can't make it happen. Well, hopefully you can find some suppliers you can settle down with and live happily ever after As someone who's long dreamed of owning a robotic camera control arm, but who doesn't have a spare $50K kicking around to buy one, I've been following the development of these kinds of projects with great interest. While this particular arm doesn't look like it would have enough payload capacity or smooth enough motion for the use cases I have in mind, the fact its a couple hundred bucks means something that does what I need it to do for an actually affordable price isn't likely too far off. Do I understand correctly that chess-moving demo decomposes into: - you recorded precise arm-movement using leader-arm - for each combination of source- and target- receptacles/board-positions (looking at the shim visible in the video, which I assume ensures the exact relative position of the arm and chess-board); - the recorded trajectories are then exposed as MCP-based functions? Bought the kit. Thank you for the great price! Are table-clamps included? Where can i find the specs. I am actively working on some project with robot arm and found following appealing eventhough this doesnt include servo or cameras or controllers. https://www.aliexpress.us/item/3256808789646447.html?spm=a2g... Interesting. Their servos seem to be PWM servos, which are available at a very good price. I would look into how to hook up all the servos—you’ll probably need an MCU to convert USB to PWM for each servo. I'm down to buy the kit and build but need some idea for how long it takes. Like, I was able to detail finish and assemble a 3D violin but could not make the time and space to assemble a full 3D printer and had to sell it. Would you please provide more info on what's involved for the kit? Ranges are okay. Great question. I would say that the first time I assembled a SO-101, it took me about 3 hours to assemble and calibrate everything. This product is largely based on the SO-101. With all the improved designs, let's say it would still take 2 to 4 hours, depending on your experience. Interesting - I was just thinking the other day that a well implemented MCP server driving a robot with access to a camera could be a really interesting project. Ordered. This is so cool. I also started looking at LeKiwi... I think I'm going to have to figure out how to make this thing mobile. Neat! Does this work with open source models like pi0 and OpenVLA? How does the inference-time teaching you outline work exactly? The software stack is built around LeRobot. So anything you can run with LeRobot should be able to run with our software. Will do more testing before the official release. Personally, I feel GR00T N1 or ACT is much easier to train and do a fairly good job I love the arm/typewriter "printer"! It's not exactly on topic (other than fun ideas, begetting fun ideas), but a USB-C/WiFi driven typewriter would be a hoot. EDIT:
Found [0] And for the reverse ... boom! (click! clack!) [1] [0] https://www.nutsvolts.com/magazine/article/turn-a-typewriter... Hey Charles, annoyed I missed out on the first batch, signed up to the newsletter looking forward to the next one! I thought your product page could use a slightly nicer UI. - I'm building an app that let's people spin up multiple variations of their pages and easily implement new UIs. - I like to put HN websites through it whilst I'm training it up to see if I can improve them. here's what my app came up with for your site: https://streamable.com/vbby9q If you want the html + css, it's here free of charge, I've split each one up with a ## Variation 1/2.. etc.. just let me know what you think - https://pastebin.com/WGNieVmq I looked at your video of alternatives and number 5 looked pretty good to me. Though, a better image of the product itself seems like the lowest hanging fruit for improving the landing. :) For stuff like this, it would be cool if you had a hosted demo of what you clicked through in the video. I'm curious about how the perception works, how do you find correspondences between the arm camera and the stationary camera ? Alright, watching the video - I'm sold, even at a sped up rate. How do I buy? I'll do in-town pickup if that's faster! Interesting project! Sorry if I'm out of the loop, but how exactly does the MCP server hand off visual data to an external LLM service to formulate the robot control actions? It's an interesting concept, but I'm having a hard time wrapping my head around how it works, because I thought MCP was text-oriented. If I'm not mistaken, the idea is use MCP to let a user-facing LLM make tool calls to a VLA model with actions the user prescribes. He mentions using the LeRobot library in another comment. +1 to this. Curious how the MCP manages base 64 image-related data and the encoding + decoding. Funny, I was just about to build an SO-101, but tariffs adding $100 to the price of the servos annoyed me. How do I buy your kit, please? Sold out unfortunately. When do you think you’ll restock? I need to ship the current orders before taking the next batch, so it will probably be sometime in late July. You can sign up for an email notification for the next batch here: https://vassarrobotics.com/newsletter Your price is too low. You should raise your price until you aren't selling out. I’m running on a very thin margin here. I want to get the robot into the hands of robot lovers without them having to eat McDonald’s all month just to save up for it. (I did that once myself, and I strongly advise against it—it’s really bad for your health, both physically and mentally) When you're selling a physical product at the single-digit-hundreds scale, "thin margin" is a false economy at best, fatal self-sabotage at worst. You should have a thick margin, which you reinvest into scaling up. The one day I was too busy at work... this came out. Darn. The price point is crazy good. If it indeed can be so flexible in learning a number of things, just take my wallet. Having an extra hand for N number of DIY projects is invaluable. The robot is only as good as the amount of time you’re willing to spend teaching it. If you put in, say, 20,000 hours and make "some" hardware upgrades, you can get it to do things like this: https://www.physicalintelligence.company/blog/pi05 I've been looking for a robot arm for a while to play ping pong with me. Curious: 1. Would we be able to control it deterministically via an API, rather than relying on LLM? 2. What is the latency on this? Do you think it would be fast enough in deterministic mode to play ping pong? Kuka had this demo years ago: https://youtu.be/tIIJME8-au8 Nowadays, if you’re running a 2–3B parameter foundation model on a standard computer, you’ll probably get a rollout frequency of 1 to 10 Hz. ACT would be faster, but I still don’t think it’s fast enough. Nothing fast enough k play ping pong is easily fast enough to cause fatal injury. That’s stick it ina. Cage with a lockout territory. High speed robots are not toys. Nothing fast enough k play ping pong is easily fast enough to cause fatal injury. That’s stick it ina. Cage with a lockout. I'm curious, since it's a YC/VC company - what's the business plan/model/vision? I assume it's not selling robot arms for $219? (Please correct me if I'm wrong!) Dalton Caldwell: “Make a product people love; worry about making money later.” Great product, congrats on the quick sales! Big fan of robotics for a long time myself. If you're in need of some website development, I'd love to chat >>hello at joshmleslie dot com<< Love it! I've been looking for an excuse to dive into AI planning for robotics, and this looks like it will make it easy to get started. Just one question: does the power supply have a 220/240v option (I'm in Australia)? Very good question. Currently, I’ve only sourced power supplies that meet US standards. However, I’m confident I can provide a power supply that works for Australia as well. When you are placing the order, make sure you put the shipping address to Australia address and so that I will know I need to source corresponding power supply. For what it's worth, I wasn't able to pick an address outside of the United States when it still had units for sale. If you want a first international sale, I'm estsauver at gmail dot com and would be thrilled to be your first customer from The Netherlands. Congratulations on a great launch! I wonder if I can strap this to my Roborock from 2020 and train it to pick up socks. Roborock sells a new model that does this [1] but it costs $3,000 and I refuse to pay that on principle when I know it's likely a straightforward model with some unsupervised training. Also I can probably fix it easier once it (definitely) breaks at some point due to collisions. I've been looking for a cheap 7 DOF arm. The only reason I haven't bought SO100/101 yet is that it's 6 DOF (and that delivery to Europe is hard to find..) I hear you, man. After using 7-DOF arms, there’s no going back to 6 DOF. I’ll try to develop a version and hopefully get it out in August. This is awesome. Can I order one where I 3d print the in printable bits and you supply the rest. Not to save money but to have a more tweakable design. We’ll probably have an electronics-only kit available in late July. However, it won’t be much cheaper, as we’re running on a very thin margin. Having built the SO-ARM 100, i feel this post is missing a lot of context for those who have not yet looked at Hugginface's LeRobot project. The most important link to get started is probably https://github.com/huggingface/lerobot How backportable are the upgrades? If I have an SO-101, can I just replace a few parts to mount a camera and use your software? The software is fully compatible with LeRobot's repo. As long as your dataset matches your inference time setup, you are good to go. I love that you're open sourcing the design! Would be curious to hear your experience as you build and sell these. I am seeing different prices everywhere:
- $199 https://shop.vassarrobotics.com/
- $219 https://shop.vassarrobotics.com/products/navrim-robot-that-l...
- $599 https://www.youtube.com/watch?v=TbDTCwzFeIU Which one is the actual price? $219 is for the unassembled version
$299 is for the assembled version
$199 is for the first 20 units of the assembled version
$599 was the price before I spent hard hours dealing with supply chains When I was just a hobbyist, I had to pay the price on the website. Now that I have some funding to order in large quantities, prices come down a lot. I do the dirty work of sourcing the components so hobbyists don't have to ;) $199 was the Founder's edition, already sold out. $219 is the unassembled version. $299 is the assembled version. Is the longer term plan for industrial robots? Would be cool if you could put other end effectors on it. As another robot hobbyist, I wish there were more detailed documentation on how things work. So many projects online just show a working demo—usually on YouTube—and it's impossible to decipher what’s actually happening, or if the robot is simply following some predefined movements. As a related question, is there some way to buy or make plush toys that are robots underneath? I wanted to use a computer to train them and download the program onto the toy, or use wifi to send the telemetry and animate it. The toy would have a microphone, speaker etc. Can this be coupled with some kind of vision AI to open/close my doggie door when pup-pup wants to? if this can be extended to full size doors this could be a very good business, no more need to modify doors to have doggy doors Opening a door is still a challenging problem for robots. Better to have electronic control of doors that can be opened programmatically. The reality is nobody can afford houses anymore, everyone is renting, so modifying doors is usually not an option. If you can work with existing infrastructure without modifying it, that is marketable. This is super cool! What have been your favorite and least favorite things about open source robots? Favorite: git clone and start using (or spending a whole afternoon to fix environment issues)
Least Favorite: some cool projects (especially in academia)—the results need H100s to replicate... I love the idea of a trainable robot arm as a learning device at that price point. However, seeing the chess demo instantly makes me think of that horrible tragedy with the robotic arm breaking a kid's finger. How strong is this to be used around kids? Yeah, safety is an important aspect. The good news is that the servos are not that powerful. Peaking at 3 Nm, with a moment arm of 0.2 m, you get 15 N of force, which is basically equivalent to the weight of three 500mL water bottles. This force might cause some scratches but should not lead to serious injuries. Initially, I was planning to launch a product using Piper Arms (much more powerful than the current product). But after testing them, I realized they could cause serious injuries if not used properly. So I canceled that version. I still have 8 of them sitting in my office. Just to check - do you ship from the USA? (international is confusing these days...) congrats on the launch. i started selling robots for $100/each back in 2012. (tapster). realized over time, it's hard to stay in business at that price point if that's all you're selling. my bots are now closer to $10K/each. you can probably keep the price low if you have some other part of the business model to fill the gap (consulting or some hosted service or an ai/data play). since you're a yc company, i assume that's the case. This is great! Given demand, I would consider a kickstarter for upcoming batches! Thank you for sharing this. Please let me know if/when it’s back in stock. You can sign up for email notifications here: https://vassarrobotics.com/newsletter It will probably be available in late July. We’re expanding the team, so we might be able to offer it sooner. Already sold out :( Any idea when the unassembled kits will be back in stock? First world problems: can I email you and specify a later shipping date? Sure! You can email us at hello@vassarrobotics.com As long as it’s within this year, I’m sure we can arrange it. So cool! I would easily pay $1000-$1500 if you put two of these on a wheel base and made it all structurally sound. Extra points if the arms sit at least 1-2 feet of the ground and can reach the ground. What does it take to fold my laundry? Not a joke. I'll pay. For hardware, folding laundry with just one arm is really hard. So probably you will need a dual arm setup. For software, even the best policy in the industry struggles a little bit (the arms they are using is about $10k each):
- https://youtu.be/Oa19cq_MxE0
- https://www.dyna.co/research In conclusion, for hardware, it might work if you have two of the arms; for software, it's probably too hard to collect enough data to make it work. Came here to ask this. Two arms? I'll pay. We will have a two-arm version available, as well as a kit for people to convert the current version to two arms—probably in August or September. How dangerous is this? If I buy/print one of these arms, how cautious would I need to be to prevent the arm from hurting me or other adults? The servos we are using in the follower arms are capable of generating about 3 Nm of peak torque. This isn’t a lot of torque, and they can’t move very fast. If absolute safety is your top priority, you could opt for servos that use 7.4V and a low gear ratio, which would further limit their torque output. The consequence is that the arm will be significantly weaker. We're going to stop using X to refer to Spring, but I haven't figured out how to do that in the standard title format yet, so I'm punting till next year. I wonder if we should just change the format to "YC YY", e.g. "YC 25" for 2025, and stop including a letter for the batch. "(YC Spring 2025)" is too long for HN titles! 1/2/3/4? e.g., (YC 1/2025). Or dots, for extra 1337ness - 2025.1, 2025.2, etc. (YC 25.1) Huh - interesting - thank you both! Notation is a funny thing...unimportant and important at the same time I'd be interested to hear about your experience working with suppliers. How did you go about finding suppliers and haggling with them? What a unique and fun build! So curious to hear about what ways it can be programmed and used for personal projects. One thing i dont undersrand about the leader follower architecture. Can I enter the kinematics myself? I like the bit of humor at the start of the video, but the bit goes on for a bit too long really cool, hoping you're able to ship it in time, given the overwhelming demand Good point. I've limited the orders to 20 units for shipping in June and 100 units for shipping in mid July. For anyone who missed out, you can find online shops that sell similar kits. Can you compare your robot to Baxter? I'm curious to see how this works. Congrats on shipping! Do you mean that Baxter by Rethink Robotics? https://en.wikipedia.org/wiki/Baxter_(robot) Thank you! How can I order it to Europe? Which country are you in? I can set up the delivery option for your country. I guess it’s too late now, please consider EU next time. Whether I would buy it would depend on total cost. I joined the waitlist. can you have two arms work together? I can personally not solder, but I would love to have two arms that can just do it for me. Need more videos, especially at 1x speed. This one shows the chess bit at 1x speed (at least I think so): https://www.youtube.com/watch?v=TbDTCwzFeIU#t=52 (I timestamped it to skip the initial voice dialog with the robot but that might be entertaining for some people as well) combine one of these with an automated robotic mower and have it pick weeds, then we are in business baby! I was searching for something like this just this week, its sold out currently, could you set up waitlist email when you can ship more please. You probably already saw this but just in case not: https://vassarrobotics.com/newsletter Post a demo without 30sec youtube ad I came to say this as well. I don't even know what the arm looks like because I shut down the video so fast after seeing that ad. Ordered. I haven't even watched the video. Let's see how this goes. Lol the website is pretty bad....could use a lot friendlier buttons, layout, more pictures, maybe some videos Sometimes a lack of solid product photos/specs/etc is a feature. (To the seller.) Please don't call it a root if it uses cheap servos
charleszyong - 6 days ago
dtnewman - 6 days ago
charleszyong - 6 days ago
loxias - 7 days ago
charleszyong - 7 days ago
tdeck - 7 days ago
amelius - 6 days ago
tdeck - 6 days ago
amelius - 6 days ago
6510 - 7 days ago
nine_k - 7 days ago
6510 - 5 days ago
nine_k - 5 days ago
6510 - 4 days ago
numpad0 - 6 days ago
6510 - 6 days ago
charleszyong - 6 days ago
slow_typist - 6 days ago
taneq - 6 days ago
charleszyong - 6 days ago
taneq - 4 days ago
fragmede - 7 days ago
charleszyong - 7 days ago
tdeck - 7 days ago
TylerE - 7 days ago
ericd - 6 days ago
brookst - 6 days ago
ericd - 6 days ago
fragmede - 6 days ago
ericd - 6 days ago
rancar2 - 6 days ago
charleszyong - 6 days ago
DidYaWipe - 6 days ago
charleszyong - 6 days ago
bertm - 5 days ago
GlenTheMachine - 7 days ago
charleszyong - 7 days ago
charleszyong - 7 days ago
GlenTheMachine - 7 days ago
You need at least six non redundant DOF to arbitrarily position the end effector in space, three for x-y-z translation and an additional three for roll-pitch-yaw. For research grade arms, I typically want at least a 7 DOF arm, which gives you a lot of cool abilities, most importantly the ability to work around kinematic singularities, and makes the inverse kinematics problem nontrivial in interesting ways. I understand you're hitting a price point, and each additional DOF costs money. I personally would pay for additional DOF. Maybe a modular design? • Adding one DOF to match ARX kinematics is doable, with a price increase of $30–40.
Yeah, typically with industrial tool changers there are spring loaded pins on the tool changer that hit pads or insert into sockets on the tool side. There will also typically be a ball detent for positive locking that is driven by a motor in the end effector. But even just a passive mounting plate and a documented connector interface would be huge. • A tool changer is a great suggestion. A few of my friends are working on kinematic couplings, which would be ideal for this. I’ll need to give some thought to how to pass electrical signals and power to the tool, while also keeping it lightweight.
You take what you can get with encoders. Ideally, you want an encoder that uses grey code, so it always knows exactly where it is no matter what. But for cost reasons this is rarely done, and you get what is essentially a relative encoder and you have to count the steps. The reason the former is preferable is that it doesn't rely on the microcontroller keeping up with the encoder, so there's no issue if you miss counts. But, again, those are as far as I know a significant step up in cost. • Could you share what functionality you want in terms of encoders? The ST3215 uses 12-bit magnetic encoders, which can retain position after power loss. Are you looking for higher resolution? For torque sensing, if the order volume is large, I can add this for just a $20-30 price increase.
Yes, but even for picking up rigid objects this turns out to be very useful. If you're picking up an eg, you want to exert a controlled positive grip force that's big enough so you don't slip but not so big that you crack the egg. If you're picking up a bolt, you definitely won't break it but many robots are strong enough to deform the threads. If you're picking up something slippery, it would be great to try to detect the slip by touch. And so on. Often, you don't know exactly how big the object is or how flexible/brittle it is and it's hard to judge by vision alone whether the fingers are even in contact with it, or if they are how much it's being deformed, so being able to control grip force is very useful. Add force and position sensing to the grippers and you can judge how deformable the object is and make decisions accordingly. • Finger tip force sensing: Is this for applications like picking up an egg?
GlenTheMachine - 7 days ago
dimitry12 - 6 days ago
liotier - 6 days ago
15155 - 6 days ago
charleszyong - 6 days ago
michaelt - 7 days ago
charleszyong - 7 days ago
GlenTheMachine - 7 days ago
clw8 - 6 days ago
softservo - 6 days ago
yardie - 7 days ago
charleszyong - 7 days ago
icedrift - 7 days ago
throwawaybbq1 - 6 days ago
guywithahat - 7 days ago
charleszyong - 7 days ago
guywithahat - 6 days ago
polishdude20 - 7 days ago
dimitry12 - 7 days ago
charleszyong - 7 days ago
fragmede - 6 days ago
dimitry12 - 7 days ago
polishdude20 - 6 days ago
victorbjorklund - 6 days ago
polishdude20 - 6 days ago
dimitry12 - 6 days ago
numpad0 - 7 days ago
notfed - 7 days ago
numpad0 - 6 days ago
GordonS - 7 days ago
charleszyong - 7 days ago
GordonS - 7 days ago
bathMarm0t - 7 days ago
euroderf - 6 days ago
bri3d - 7 days ago
bittercynic - 7 days ago
charleszyong - 7 days ago
charleszyong - 7 days ago
aesch - 7 days ago
charleszyong - 7 days ago
chrishare - 7 days ago
rfwhyte - 7 days ago
dimitry12 - 7 days ago
sabareesh - 6 days ago
charleszyong - 6 days ago
bredren - 7 days ago
charleszyong - 7 days ago
iamflimflam1 - 7 days ago
pbrb - 7 days ago
martythemaniak - 7 days ago
charleszyong - 7 days ago
Nevermark - 7 days ago
davidweatherall - 6 days ago
bredren - 6 days ago
quadrature - 7 days ago
laidoffamazon - 7 days ago
peepeepoopoo135 - 7 days ago
bconsta - 7 days ago
codybontecou - 7 days ago
timmg - 7 days ago
pachevjoseph - 6 days ago
charleszyong - 6 days ago
jimrandomh - 6 days ago
charleszyong - 6 days ago
jimrandomh - 6 days ago
didip - 6 days ago
charleszyong - 6 days ago
TuringNYC - 7 days ago
charleszyong - 6 days ago
TylerE - 7 days ago
TylerE - 7 days ago
cjblomqvist - 6 days ago
charleszyong - 6 days ago
catwhatcat - 6 days ago
drodgers - 7 days ago
charleszyong - 7 days ago
estsauver - 7 days ago
lchengify - 7 days ago
tomp - 7 days ago
charleszyong - 6 days ago
bravesoul2 - 7 days ago
charleszyong - 6 days ago
Tepix - 6 days ago
fragmede - 7 days ago
charleszyong - 7 days ago
tdeck - 7 days ago
benob - 7 days ago
charleszyong - 7 days ago
bufferoverflow - 7 days ago
bagels - 7 days ago
_tqr3 - 7 days ago
EGreg - 6 days ago
hbarka - 7 days ago
mclau157 - 7 days ago
mlboss - 7 days ago
dheera - 7 days ago
codekansas - 7 days ago
charleszyong - 7 days ago
vasusen - 7 days ago
charleszyong - 7 days ago
cadr - 7 days ago
hugs - 6 days ago
surfmike - 7 days ago
_pktm_ - 7 days ago
charleszyong - 6 days ago
ZeroCool2u - 7 days ago
pama - 7 days ago
charleszyong - 7 days ago
pj_mukh - 7 days ago
namank - 7 days ago
charleszyong - 7 days ago
norcalkc - 6 days ago
charleszyong - 6 days ago
bhelkey - 6 days ago
charleszyong - 6 days ago
altairprime - 7 days ago
dang - 6 days ago
slater - 6 days ago
altairprime - 6 days ago
dang - 6 days ago
aesch - 7 days ago
vavooom - 7 days ago
brcmthrowaway - 6 days ago
jjfoooo4 - 6 days ago
anythingworks - 7 days ago
charleszyong - 7 days ago
myth_drannon - 6 days ago
Javantea_ - 7 days ago
charleszyong - 7 days ago
polskibus - 7 days ago
charleszyong - 7 days ago
polskibus - 6 days ago
whazor - 6 days ago
fragmede - 7 days ago
dang - 7 days ago
whoomp12342 - 7 days ago
qwert12345887 - 6 days ago
dang - 6 days ago
matt3210 - 6 days ago
ModernMech - 6 days ago
DarmokJalad1701 - 7 days ago
mclau157 - 7 days ago
lysace - 7 days ago
xchip - 7 days ago
