The Sound of One Robot Arm Clapping - Dexter HDI: The Future of Robotics or Just Another Arm?
Welcome to the Robotics Roundtable
Welcome to our newsletter. In it, we take an article that piques our interests and we discuss it from our unique perspectives. Sevy - the robotics hardware engineer, Connie - the robotics software engineer, and Sean - the social entrepreneur and marketing + finance dork.
This week we selected a video that delves into the advancements in robotic arms: "Dexter HDI by Haddington Dynamics". The video, and articles published about Haddington, piqued our interest for several reasons.
Firstly, the Dexter HDI is a 3D-printed, 5-axis robot arm priced at a humble $2,999. It is a company who managed to make a good looking and low-cost robot arm by leveraging advances in 3D printed materials. The company was recently acquired by the Ocado Group.
Second, the Dexter HDI stands out with its unique FPGA supercomputer optical encoder system. The FPGA can process up to 2 million samples per second, allowing Dexter to adapt quickly to any disturbances, enhancing operational safety. It also allows Dexter's optical encoders to be placed directly on each joint, achieving high-precision movement and positional awareness that surpasses competitors, including those at much higher price points.
Third, the user-centric approach of Dexter HDI, with features like the "Follow Mode" and customizable end effectors, signifies the potential of robotics to become more accessible and intuitive, even for those without a technical background.
If you enjoy our work, please consider sharing, visiting our blog, and subscribing to our newsletter where you can read more about our thinking, get updated on our latest events, hear about breaking robotics news coming out of Los Angeles, and follow our journey as we work to unite the Los Angeles robotics ecosystem.
CONNIE’s Corner
This video comes from the Slant3D channel, a company that does 3D printing, and not from the robotics company itself. While a nice overview of Haddington's 3D printed aspect, I disagree with them when they say it's underutilized. 3D printing is an industry standard manufacturing process these days. Like any manufacturing option, it has pros and cons. It was the right choice for Haddington as they wanted something low cost, low units, low load, high complexity, and with the ability to modify their design often. However, I don't believe that their 26 million dollar acquisition price tag is due to the 3D printing aspect. I believe it's due to the optical encoder design mentioned in passing.
3D printed robot arms are fairly common in the robotics hobbyist community, but suffer from poor precision and tracking. Their FPGA optical encoder solves this problem with a crazy interpolation from the gradient of light seen by the receiver, with the ability to process the massive amount of data they get from this design. This gives them far more data than a standard binary optical encoder and enables the 0.8-1.6 million points of precision metric listed on their website. It also means that their joints are smaller, and that their stepper motors don't require other sensors. This makes Dexter cheaper to produce, easier to maintain, and in a smaller form-factor, if not quite as sleek as a UR5. The FPGA encoder is probably what most of their purchase price came from; if I were a large robotics company I'd want to snap up that tech for my own robots as well.
More than the 3D printed aspect, I think Haddington chose its feature set well. For example, I think torque repeatability is one of the coolest features - the ability to play back recorded motions is not new, but they've really streamlined the process. Combined with the precision, I can see many applications from painting to soldering that this arm could be used for. They've made it look reasonably nice for a 3D printed arm, while undercutting other cobots at just $2,999 for a kit. Compared to $23,000 for a UR3 it would seem like a no brainer to customers on a budget. Robot arms have a nice, consistent demand and I think Haddington has really carved themselves a niche in this market.
SeVY’s Corner
At the end of the video, the guy from Slant3D asks an interesting question: "Why is [3D] printing not used for products more often when it so obviously creates so many benefits that translate into real actual returns for customers, for investors, and for founders?”. Put another simpler way: why has 3D printing stayed mostly in the hobby realm and not expanded to manufactured products? Haddington Dynamics, the company mentioned in the video is a great example of an exception to understand the answer to that question.
So let us talk about the alternative method that most products employ for manufacturing plastic parts. Injection molding requires an expensive large steel mold that has to be machined precisely. However, every individual part is inexpensive because it is fast. You shoot hot plastic material in the mold, it cools, and in under half a minute, you have a part. In an eight-hour day that amounts to one worker making 1000 parts. So expensive upfront cost than low per unit cost.
So let’s go through Slant3D’s points on why 3D printing is superior:
They can improve their arm design as they build. 100% correct. I do this every day for the robot I work on. Iterating with 3D printing is invaluable but once it works then you can move to a mold because you have locked in the geometry that works. The more moving parts the more iterations of design.
They save $100,000 upfront in cost for molds. That ballpark cost is close to the real figure even accounting for combining different parts in one mold. This is true, they save the molding cost at the expense of individual parts being more expensive because they take longer to print. There is a specific number of units that is the break-even point where both processes are the same cost.
Complex geometries are easier with 3d printing. This is absolutely true. Most of the time an engineer can make the parts work within the constraints of molding design. But printing is superior for more complex shapes like sprockets and gears, which the arm has many of. A moving robotic arm has more complexity than most assemblies because several moving parts have to be hyper-accurate.
It is faster to 3d print. This depends on the amount being made just like the cost. Molds typically are 3-6 weeks and then you can start pumping out parts quickly. So to get thousands quickly molding still works as long as you have the capital. However, if you need 100 parts in a week 3D printing is the only real option.
So why did 3D printing manufacturing work for Haddington Dynamics? I think it is because:
They started on Kickstarter with a low number of sales.
They had to get those sales to customers quickly and for low upfront costs.
They had a complex part assembly which would have taken longer to convert to molded geometries and which 3D printers are good at.
They had already designed those complex parts by printing them so they knew they would work.
Also, their customers were the hobbyist who already loved 3d printing and could make easy add-ons or replace faulty parts. Huntington could generate free cash flow from selling print files that they already had. It made a lot of sense.
So let’s go back to the question: why has 3D printing stayed mostly in the hobby realm and not expanded to manufactured products? It’s because 3D printing manufacturing is slower and more expensive when making a lot of a product. What will flip this equation is if printing time decreases substantially without adding additional steps. For now, 3d printing is great for small batches of complex geometries and testing and iterating designs.
SEAN’s Corner
A robotic skeleton key
I’m rarely interested in reviewing promotional videos, but this one piqued my interest for two reasons:
First, it is the story of a successful robotics startup launched from kickstarter.
Second, the technology holds promise for a new paradigm for science.
A New Funding Model
It’s not unusual that a consumer product leverages Kickstarter, but it stood out to me that a hyper-niche, mostly B2B product like a robotic arm was successful.
Traditionally, robotics startups go the academia/venture route, but I’m hoping that this startup success gives Founders new options. An option that gives startups the option to spend more time exploring and inventing before they need to run to market - ultimately creating better and more innovative products.
A New Paradigm
What I see in the automation industry is the centralized development of high-production value, market specific robotics.
The current market and funding system make scalable robotics almost the default choice for robotics startups.
Further, it takes effort to design and assemble a robot, meaning the engineers must rely on off the shelf parts, often forcing a compromise in their design and further reinforcing the current startup focus.
However, this scalable, high-production value robotics paradigm means that long-tail robotics use cases go unserved or these use cases must fit their operations into the current paradigm.
For example, robotics applied to the acceleration of science discovery is slowed because each lab application is unique implying a small potential market. Alternatively, larger robotics systems are integrated into labs, but that means a compromise both for the humans and, since the systems are expensive, only certain research applications make financial sense. Further, the heavy investment makes pivoting research approaches a heavy choice.
However, this system, having proven the ability to produce 3D printed parts that can create high-precision robots, means labs can design their own systems, fit to their function, while keeping their costs low, without sacrificing quality.
In a previous post, we discussed the power of lab integrated robotics to radically accelerate discovery (you can read that article here: Trillions of Experiments per Year). A technology like this has the potential to similarly speed discovery in many other areas of science.
I could imagine optical experiments that self-organize, chemical experiments that self-discover, robotic cameras that collect animal behavior without direct need for human intervention.
In short, this startup could usher in a decentralized revolution in science.
Synthesis
What makes Haddington Dynamics special and successful? Sean envisioned a world where anyone could print a robot arm at home, democratizing the robotics space and finding increased usage in the science sector. Sevy and Connie both agreed that while the 3D printing was an interesting aspect of the Haddington arm, their chief technical contribution was their novel FPGA optical encoder that enables high accuracy on a traditionally low-accuracy manufacturing process. The engineers focused on how 3D printing was mostly a means to an end. 3D printing allowed Haddington to leverage novel funding platforms like Kickstarter, and allowed them to iterate quickly when they started, but does limit the size, scale, and strength of the arm.
The cost-effective nature of the arm, combined with features like torque repeatability, means it has carved itself a niche in the arm market. Sean envisions an era of automated science, where low-cost arms are used to run autonomous experiments. Sevy and Connie agree that experiments, with their repeatability and usually lightweight objects, would be a good use of the Dexter arm. In addition, they are pleased that Haddington is focusing on a feature set that improves ease of use – programming robot arms is a high barrier for entry, which is alleviated with their torque repeatability feature.
Overall, we all agree that Haddington is doing a great job with their feature set, but we will be watching with interest to see if they continue to focus so heavily on 3D printing as they scale. We will also be seeing which industry ends up adopting robot arms, and if the low price tag will spur some new markets to embrace automation.
We hope you enjoyed this blog post. If so, please share with someone who would also enjoy it.
Thanks,
Sevy, Sean, Connie