Numina had a power problem. Founded in St. Louis, Missouri, the data startup brought a thoughtful approach to smarter cities — a light pole-mounted sensor combining a camera and computer vision to meticulously count cyclists and pedestrians no matter how much they swerved. With support from the Knight Foundation, they had successfully deployed several prototypes with the aim of empowering urban planners to design streets for people instead of cars. Numina’s approach protected people’s privacy by analyzing footage within the sensor, keeping the anonymized movement data while discarding the images.
Attaching the sensor to light poles was one thing; using their electricity was another altogether. Every city’s policy is different, and the bureaucracy can be stifling. As a backup, Numina’s co-founders had soldered together a makeshift pack of D-cell batteries to supply their earliest prototypes with only a day or so of power. Their hack worked, but the duct-taped sensor — assembled in an off-the-shelf enclosure in a co-founder’s living room — was ultimately unsustainable. “It was literally a box of parts,” says CEO Tara Pham.
In the meantime, as a proof of concept, the team resolved to swap the packs frequently. Each time, they needed to carry a 12-foot step ladder down the street, block off the pole with traffic cones, and climb up to replace the sensor. Was there a better solution?
In early 2016, Numina joined the inaugural cohort of URBAN-X, a New York City startup accelerator founded by MINI, the iconic car manufacturer now part of the BMW Group. Launched out of the company’s commitment to improving city life, URBAN-X is run in partnership with the venture fund Urban Us, semi-annually soliciting startups tackling some of the toughest challenges cities face today. Together, they create disruptive and scalable solutions across sectors such as real estate and construction, infrastructure, mobility, food, water, energy, and more.
In return for a small equity stake, Numina and its fellow teams received cash, guidance, and the undivided attention of the experts-in-residence (EIRs) — an in- house team of hands-on UX and UI designers, mechanical and electrical engineers, and specialists in working with cities. The first of their kind for an urban tech accelerator, the EIRs are charged with addressing teams’ weaknesses, helping them learn the most important things first, and quickly getting their prototypes to a place where they are market-ready.
Our goal was to develop a design for manufacture, says Pham. We joined knowing we’d have BMW engineers working on our product.
One was Johan Schwind, design director at URBAN-X, and a former concept designer for both BMW and Audi. Another was Dean DiPietro, an industrial designer with twenty years of experience creating dozens of consumer products sold around the world. At Schwind’s direction, the program modeled its approach on “design thinking,” which calls for listening intently to customers, prototyping ideas quickly and cheaply, testing them, and iterating rapidly to learn as much as possible at minimal cost. (See: “The URBAN-X Design Manifesto,” right.)
Iteration is the core of the URBAN-X process. Hone product-market fit quickly in small, manageable steps, reducing risk and expenses. Oscillate between phases of divergent thinking – generating a variety of ideas — and convergent thinking, focusing on one. Then do it again.
Engaging them early is critical. Build hypotheses on their feedback and observations, then set clear, quantifiable metrics for testing. Define how positive and negative outcomes affect development — and be prepared to rethink everything.
Technology is only a means to an end. Dig deep into customer needs, identify true pain points, and choose technologies that play to your strengths (and budget).
Find, hack, and repurpose tools and materials for rapid testing rather than building prototypes from scratch. Take full advantage of open source hardware and software.
“Like many teams, Numina came with great advances in what they were good at — in this case, machine vision — and needed the most help in areas they were not, which was hardware,” says DiPietro. “Their dilemma crystallized what we do at URBAN-X — our real expertise is in turning promising technology into products solving big problems.”
With only 8 weeks until Demo Day, the EIRs huddled with Pham and co-founder Martin McGreal to define the constraints of a design that could be manufactured at scale. Because governments are less price-sensitive than consumers and more concerned with reliability, Numina’s finished product wouldn’t be constrained by cost over quality.
Three challenges quickly became apparent. The first was power — reliance on only photovoltaics remained off the table due to cost and size, and light poles inconsistently received power. They would have to crack the battery problem.
The second issue, related to the first, was running hot and cold. Any enclosure they selected would have to dissipate heat from the processor while protecting the sensor against any and all weather, including rain or snow. Their benchmark would be a NEMA 4X or IP56 rating, meaning the case must be water-and dust-tight.
Finally, it would have to stand out. Numina doesn’t hide the fact it collects anonymized data, and it doesn’t want to be confused with standard-issue surveillance devices, either. “We didn’t want to make a gray, inconspicuous box,” says Pham. “In Downtown Brooklyn, our sensors are bright yellow, as if to say ‘this is not a secret we’re collecting data.’ This is community data, collected by Numina.”
Ultimately, the three challenges were entwined — the battery’s size and composition would define the shape and dimensions of the enclosure. Numina’s team selected a lithium iron phosphate battery, which has several advantages over more conventional lithium ion batteries — including a comparable energy density (around 220 W/I) while offering significantly higher thermal stability and safety. (In other words, they cost less and were less likely to burst into flames.)
With one problem solved, the team turned to the next. The new batteries’ sheer size presented an irreducible design constraint dictating their next move — how best to package the batteries, camera, and processor in a single enclosure. Schwind’s original suggestion was to simply purchase one meeting their specifications.
“We tell so many companies, ‘What are you doing? In this case, you’re a machine vision company, not a hardware company. Buy something off the shelf,’” Schwind says.
But this time around, a traditional camera enclosure “would have cost a thousand dollars just to start — and that’s not including development costs.”
By contrast, the team had set an optimistic internal target of $100 per unit for material costs. Hemmed in, they brainstormed and debated over available options. “It couldn’t be metal because that would block the signal,” DiPietro explained. (Numina’s sensors transmit their metadata via cellular networks.) “Somehow, it had to be plastic.” And because of their expense, the team also wanted the option of adding or subtracting batteries as necessary — which meant a modular design.
At a loss on how to proceed, the group tabled the discussion. During that interval,
inspiration struck. “I think it was Johan, or it might have been Dean,” Pham recalled, “but one of them said: ‘Really, this should just be a tube — or a pipe.”
Instead of paying hundreds of dollars for an aluminum box, they could spend pennies- per-foot on PVC pipe, one of the world’s least expensive industrial materials. But this brought tradeoffs, too. Cutting and drilling to fashion an enclosure required time and experimentation as well as the appropriate tools.
“Our job became working with an affordable, off-the-shelf material, and then dealing with the consequences of it being affordable, which is irregular tolerances”, says DiPietro. “We were purchasing inexactness at low cost.”
As the EIRs worked with Numina to develop a prototype, the question of scale crept into the conversation. “It was very clear this would work for one sensor, but would it work for thousands of them?” Schwind remembered thinking.
Even using PVC pipe, it would have been an expensive mistake to move too quickly into injection molding, which is cost- effective only in mass quantities. By keeping the ultimate design of the sensor digital, in CAD software, they could tinker with parts that could alternately be casted or printed, offering greater flexibility in working with low-volume manufacturers.
For example, while the pipe would contain the processor and battery housing, how would they expose the camera and cap the enclosure with sufficient waterproofing? An off-the-shelf rubber cap was summarily rejected because “it made the whole thing look like it crawled out of the sewers,” says Schwind.
Casting about for in-house alternatives, the EIRs settled on using a pair of 3D printers at URBAN-X to print caps integrating the cameras inconspicuously. “Printing isn’t the most exact process, but it’s more exact than these drainage pipes were,” says DiPietro. To ensure compliance between the precisely manufactured caps and lower tolerance pipe, the pair added rubber gaskets, which doubled as weatherproofing.