Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Berkeley Blue – Accelerating robotics through low-cost hardware (berkeleyopenarms.org)
105 points by hugs on April 13, 2019 | hide | past | favorite | 49 comments


Very, interesting project! I found it so inspiring that I included it in the Weekly Robotics #34 that is live since 30s ago (https://weeklyrobotics.com/weekly-robotics-34).

For anyone interested here is a link to the paper: https://arxiv.org/abs/1904.03815 . Having robotic arms for <$5k would be great for research, hobbyist and applications not requiring crazy accuracy. I don't have hard data at hand but I doubt that at present you can find commercially available robot with similar characteristics for less than $15k.

Another thing that's happening here and I really enjoy is how BLDC motors are being used in applications like these. The reason I like it so much is because hopefully there will be a market for cheap actuators that will do the job well enough to be used for some applications. I'm pretty sure that among other things I've seen a lot of BLDC based actuators on quadruped robots.


There are already quality German robot kits for 5k: https://de.rs-online.com/mobile/p/roboterarm-montagekits/176...


The specs are much lower - you get only 5DOF for $5,600, instead of 7DOF. Igus is working on their ReBeL arm, which may be a more competitive option.


Is there any specific application, that needs 7DOF and cannot be done any other way? I guess, that Igus will give you other price when you ask them directly and 50 pieces. I am not related to them, but I like their approach. Plus I have no problem controlling stepper motors directly.


Are there any "hobbyists" who can afford a robot arm at $5k?


I'm sure there are some. Also if this robot arm with 7 DoF costs $5k, where about 70% of the costs are related to motors and electronics then how much would a 4 DoF arm cost?


Should be plenty. 5k is not a lot.


Abstract— Robots must cost less and be force-controlled to enable widespread, safe deployment in unconstrained human environments. We propose Quasi-Direct Drive actuation as a capable paradigm for robotic force-controlled manipulation in human environments at low-cost. Our prototype - Blue - is a human scale 7 Degree of Freedom arm with 2kg payload. Blue can cost less than $5000. We show that Blue has dynamic properties that meet or exceed the needs of human operators: the robot has a nominal position-control bandwidth of 7.5Hz and repeatability within 4mm. We demonstrate a Virtual Reality based interface that can be used as a method for telepresence and collecting robot training demonstrations. Manufacturability, scaling, and potential use-cases for the Blue system are also addressed. Videos and additional information can be found online at berkeleyopenarms.github.io.


Looks like some great hardware! I love the simple joint design and the variable impedance. I hope sales are successful and this venture is able to perpetuate itself!


Yes. The drone and drone camera industry has made 3-phase brushless DC servomotors with position encoders into consumer products. This robot uses camera gimbal motors. Most previous robots in this size range used either more expensive industrial motors or variations on R/C servos like the Dynamixel.

There are many Dynamixel based robots. It's basically an R/C servo with an encoder and a useful communications protocol. It's a fast motor with a lot of gear reduction, not very back-driveable, certainly not enough for force feedback. Typical pricing is in the $1000-$2000 range. The tier below that is robot arms with regular PWM R/C servos. I have a UArm like that on my desk. Repeatability is about 5mm. Costs a few hundred dollars. Not back-driveable at all.

Servomotors and motor controllers used to be incredibly expensive for what they do. Over $1000 for a motor controller was not unusual. The actuation end of robotics is getting cheaper, at last.


Motors from hobby market with custom controllers + belt reduction instead of precision strain wave gearboxes seem to be major cost-reducers. Less precision, but lower cost. More than good enough for perception and control research on how to perform everyday chores.


What makes this work is that the better hobby motors for drones are now pretty good. Hobby motors used to be cheap sheet metal junk.


(Brushless) Hobby motors are typically CNC's aluminum and steel these days. Some sheet metal ones exist too. High quality closed loop controllers shipping in volumes helps too. The cheap motor factories can make higher quality motors upon request too so "hobby like" motors can be great in robotics if a big buyer sources them right.


"High quality closed loop controllers shipping in volumes helps too."

Yes. Those things used to be way overpriced. I was at an industrial trade show some years ago, and saw that Maxon, a motor manufacturer, had added controllers to their product line. I was chatting with the sales rep, and he said they'd added them because they were tired of selling $100 motors that were used with $1000 controllers, when motor and controller cost about the same to make. Selling both together resulted in a huge improvement in profit margin.

Encoders, too. For a long time, shaft encoders, which are very simple devices, were way overpriced. They were separate from motors, which meant you usually needed a motor with a shaft out each end, plus the parts to attach motor and shaft, plus the alignment problems of keeping two things with separate bearings concentric. With the encoder built into the motor, this all gets much simpler and neater.


Seriously. I dream of hobby motors with built in high resolution encoders in the sub $100 price range. That and a programmable closed loop brushless controller for under $50 would be so impactful.


Define low-cost? What sensors are being used in the video?


Exactly... There are a lot of open source robot arms on thingiverse and youtube. Some of them are $100 dollar proof of concepts with basic RC servos [1] to full blown awesomeness like the BCN arm [2] with many flavors in between.

It would be super cool to get the prices of the control boards down, standardize the software stacks that run these things, and really integrate them into everyday life. For now, I suppose we have to watch a group of people reinvent the ARM every few months.

[1]http://www.eezyrobots.it/eba_mk2.html [2] https://www.bcn3dtechnologies.com/en/bcn3d-moveo-the-future-...


According to the paper, "can be <$5000 if a run of >1500 is made"


That seems ridiculously expensive. How much is a Universal Robotics "Co-Bot" or one of Kuka's lower end models?


The startup I work at (Automata) is selling a robot arm for £5k ($6.5k). Getting the price that low took a lot of work and there's definitely mechanical tradeoffs to get there.

Robot arms are pretty complicated electromechanical devices with high torque requirements. I suspect one day the manufacture of them will follow a similar path to cars, but until then building them cheaply will be hard. The volumes just aren't high enough yet.


There are a ton of industrial robot arms available in China in the last few years. Examples: https://item.taobao.com/item.htm?id=572669195615 @ USD$3550, https://item.taobao.com/item.htm?id=588690121022 @ USD$3850, https://item.taobao.com/item.htm?id=572700770567 @ USD$4100, https://item.taobao.com/item.htm?id=589952267004 @ USD$4251, https://item.taobao.com/item.htm?id=582638257935 @ USD$10000, etc.

Some advertise using imported SMC (Japanese) pneumatic/hydraulic mechanisms.

This may not be unrelated to the fact that you can buy a fully detailed product model like https://item.taobao.com/item.htm?526498275889 or https://item.taobao.com/item.htm?id=571064427455 for USD$0.30+.


There is one important difference: article's robot is human-safe.


Sure but simple presence / proximity type sensors are cheap. As are cameras. Human-safe is then basically a software feature.

IMHO the whole 'cobot' media fad and deployments with humans are needlessly complex versus just handing over between dedicated human-only and robot-only motion envelopes.

In industrial or small-scale automation settings, useful for a subset of tasks at best. For these special cobot cases, why not use a retrofit kit for third party arms? It makes more sense as would be more loosely coupled with inevitable hardware evolution and provenance.

Hell, it'd even work with arbitrary industrial gear. Talk about a large retrofit market.

Software people: that's a valid (and high fad-factor!) startup idea. Get some VC and go get 'em.


It may work for Chinese buyers, but it's a tough sell outside of China. What about manuals, software, tech support?


I'd assume they probably clone an established interface allowing third party PLCs / software to be used.


No it's not. Try to find another 7 DOF arm with 2 kg payload and similar envelope for less than $5K.


Might be wrong, Co-Bots start at $35k and Kukas are custom-built, often get over 6 figures.


It seems to be on the very cheap end if it would qualify as an "industrial robotic arm" (I don't know if it does). I found this paper from 2012 [1] with a list of "Low price industrial robotic arms" (Table 2) and the cheapest one there is $17k.

[1] https://www.researchgate.net/publication/241632269_A_brief_s...


Not Industrial...

"During a 7 unit fabrication run in-house, Bill of Materials (BoM) cost for each arm was tracked at $3328 as shown in Table II. Plastics were printed on Markforged Onyx One and Monoprice MP Select Mini 3D printers. About 75% of our prototype costs were motors and driver boards."


What's not industrial about that quote? That they used 3D printing to prototype?

That said, they seem to aim primarily at labs, not factory floors.



I was so disappointed when I found out that thing is just a robot arm that takes a cup from a traditional coffee vending machine and puts it in front of the customer window. It seems weird that we don't have a robot system that can make a coffee using a traditional espresso machine.


It's not even a traditional vending machine. It's three super automatic espresso machines, hooked up to probably an arduino or something. A 1970s coffee vending machine would be cheaper.


Just to add in, the lowest priced “industrial” arm projected prior to this was the “Franks” robot, where they anticipated to sell them at quantity for about $10k. Note the current development version of that arm is at least 2X that price.


This isn't "industrial"...

"During a 7 unit fabrication run in-house, Bill of Materials (BoM) cost for each arm was tracked at $3328 as shown in Table II. Plastics were printed on Markforged Onyx One and Monoprice MP Select Mini 3D printers. About 75% of our prototype costs were motors and driver boards."


Cobots are a multiple of that, and even a lot simpler arms easily go for 10k or more.


What is the expected lifetime?


Awesome. But I think the real revolution in robotics is when tact can be built in throughout the arm. Feeling force feedback in each joint is one part of the equation; but tact is the way that subtle manipulation, precise feedback and deep environment status understanding will close the gap with human manipulation.


Berkely also has some interesting research on embodied AI.

My belief about robotics is that we need really efficient and powerful artificial muscles and closer biomimicry. It seems like the way real limbs operate with leverage will be an advantage if we can get useful muscles.


Are there any DIY robotic arms that you would recommend below $1000 ?


Depends on your payload and complexity of what you want. There are a couple of design on Thingiverse that uses hobbyist RC servos. The total cost should be less than $100.


Can you point me to them ?


They can range expensive to around $200 [1] that is like Universal Robotics clone to really cheap using hobby servo that cost less than $30 [2]

[1]https://www.thingiverse.com/thing:3327968

[2] https://www.thingiverse.com/thing:360108


Are there any prebuilt ones?


Yeah, they're a couple you can buy online [1]. Dobot came out one that looked nice and the mechanical spec sounds pretty decent [2]. But I hear the software is shitty so YMMV. They recently came out with an Arduino "AI" library so not sure if that's easier to program. If you want to go cheaper, Lynxmotion is the next one on the list. If you want to go super cheap, the LittleArm series is the way to go. I remember following them on Kickstarter and the owner (it's like a two man team) released a GUI that looked useful. It's coded in Processing which allows you to record the arm movement and playback.

[1] https://www.robotshop.com/en/robotic-arms.html [2] https://www.dobot.cc/dobot-magician/product-overview.html


Thank you!


In the video, the robot is remotely controlled.

How long will it take until reinforcement learning / ML can do these things automatically?


Will robotics be 2050s version of 1950s kitchen appliances ?


Is there a MuJoCo model?




Consider applying for YC's Summer 2026 batch! Applications are open till May 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: