Sensor Fusion: Reinventing Retail’s Store of the Future, Today
Since the beginning of retail, store managers have counted front door traffic, first very manually, then more systemically through beam counters and other technologies. The most accurate way to count store traffic has long been through video analytics, and there are a variety of high-quality, fairly inexpensive stereo video cameras available from manufacturers around the world to fill that need.
However, as physical stores attempted to keep pace with online channels powered by their array of advanced website analytics, retailers demanded data streams from a variety of additional sensor technologies—Wi-Fi and Bluetooth Low Energy (BLE) devices among them—which led to additional hardware investments, installation and deployment costs, and a messy, cluttered ceiling of sensors.
What was missing was a comprehensive, all-in-one sensor integrating video, Wi-Fi, and Bluetooth, and designed specifically for the complex needs of the retail industry. It was exactly that unmet need that led RetailNext, a SaaS software company, to design, manufacture and ship its first-ever hardware device. Doing so successfully happened despite a one-year timeline to delivery and both personnel and financial constraints.
Most every project team is well familiar with the Project Management Triangle built around scope/features, cost and schedule. The general rule of thumb for stakeholders is to “pick any two.”
It makes for a memorable sound bite, but what if business requires you to pick all three? Faced with this challenge, I’ve highlighted my suggestions for realizing the improbable, on specification, on budget and on time, developed through my role in bringing an all-in-one sensor to market.
A “Blank Slate” Doesn’t Preclude a “Head Start”
Because it would be a world’s first, the project team was afforded the opportunity of starting with a blank slate. However, the short timeline didn’t necessarily allow for the luxury of a completely blank slate.
Rather, the project team defined its performance parameters and searched for comparable existing platforms. The team knew it needed to integrate multiple hardware components—including two video cameras for stereo video—into a single box, with high performance processing capabilities for current (known) and future (unknown) applications.
Those product requirements sounded an awful lot like … today’s mobile smartphones.
That realization led to the development of the Aurora on the Qualcomm® Snapdragon™ 805, maximizing the sensor’s video capture and output capabilities, as well as its processing efficiency.
The existing development platform allowed the project team to add and take away—in a custom fashion—exactly what it needed for its purposes, and the processor’s existing encoder, GPU and DSP were to prove valuable for the kind of analytics RetailNext customers would be running on the new sensor.
Starting with an existing platform made the tight timeline not only feasible but even reasonable.
Parallel Paths Compress the Timeline
Building on the foundation of the Qualcomm processor delivered additional benefits when it came to developing the sensor’s software. In scoping the original project, software development was anticipated to be the long pole, and in the end it proved to be so. However, by procuring a bunch of existing development kits from Qualcomm, software development started in parallel to the hardware design. As a result, several months were effectively cut off at the end of the timeline, allowing the entire process to be completed within our customers’ scheduling requirements.
An Expanded Bench Strengthens Capabilities
In getting started on the project, one of the first bits of advice from the board was, “keep your project team small.” Doing so allows for direct line of sight and ease of communication, both of which help foster enhanced teamwork, collaboration and follow-through.
But, there was still the timeline, and to make sure we kept our commitments, we were required to take advantage of third parties and their massive teams of collective expertise and experience. Within Qualcomm’s embedded business model, we had project-ready affiliates in manufacturing set up from the very beginning, allowing the contract manufacture of quality hardware in a minimal amount of time.
Very early in my career, a mentor once advised, “The most difficult challenge you’re going to encounter is when you transfer your North American manufacturing prototypes to overseas contractors.” Sound advice at the time, but with worldwide alliances and the relative ease of global commerce today, the Aurora project team never had to worry about that.
Manage Feature Creep
Feature creep is an almost inevitable threat to any project. Some companies approach those impending threats with timelines and formal drop-dead dates that mandate no changes. While effective at eliminating feature creep, the unintended consequences often include a finished product that doesn’t meet the end customers’ needs.
During the Aurora project, the team was regularly approached with modified requirements, like the need for the sensor to be adaptable to taller ceiling heights. Our approach to these requests was, “The answer is never ‘no,’ but let me tell you of the tradeoffs that will have to happen.” A collaborative discussion about tradeoffs like schedule delays and increased costs allowed for better decision-making all around.
The nature of today’s business climate often presents tremendous opportunities, yet ones coupled with sometimes daunting challenges. Learning to maximize your internal resources while taking advantage of available external resources can help you not only turn the impossible into the possible, but even to the probable.
Bill Adamec is the Director of Hardware at RetailNext, where he leads research and development, engineering and manufacturing teams on the design, development and deployment of hardware infrastructures for retail analytics applications.