The world of consumer products is full of new and innovative technologies, from ever smarter phones to self-driving cars. Companies creating consumer products feel the constant push to incorporate innovative functions, usually countered by the need to sell at an acceptable price. Innovation at near-zero incremental materials cost is easier for software apps for mobile phones, because electrons are free, but it’s harder when making a real product. This balancing act requires a thorough knowledge of where a market is going and finding a cost-effective route to provide what the market wants.
Take consumer lawn mowers as an example of where innovative functions, and even service features, might be added to consumer products. The top end of the market offers robotic devices that trundle around your grassy areas like well-trained Roombas. Despite the high price on the box, these machines were designed to meet a surprisingly low bill of materials cost. The intelligence, such as it is, is added by asking – forcing – the consumer to install a boundary wire around the lawn, which the robot mower then uses to detect the edges.
In an ideal world, however, intelligent lawn mowers would not force their owners to undertake additional installations when they come out of the box. Indeed, this is a challenge for the next generation of high-end robotic mowers, a new function that could be made possible by machine learning. In commercial mowers for trimming golf courses, meeting this challenge is relatively straightforward. All it takes is a big computer, cameras and GPS and you’re done. For the consumer product manufacturer, the challenge is greater – every dollar extra spent on processing could mean ten more on the eventual price tag.
The need to support additional functionality without (much) additional cost is an engineering challenge, and for lawn cutting manufacturers could take a number of possible routes. Vision-based systems are likely to become more common in commercial products because cameras, such as those found in web-cams and phones, are already cheap. The processing overheads required to process the images might be met using application-specific silicon, or newer more capable processes at the sub-five-dollar level, with appropriate algorithms.
Alternatively, cloud-based computing as a back haul may be attractive. This allows more powerful computing to do the hard work remotely, while sharing the cost among all product owners. If challenges due to wireless connectivity in large gardens could be resolved - as they surely will - there are many attractive features that make this idea compelling. As well as removing the cost of the processor from the product, the use of remote computing allows the manufacturer to implement upgrades and additional services, such as paid for add-on features. The product becomes a platform for selling additional services. Additionally, cloud-based functionality would allow the manufacturer to maintain control of their intellectual property. It’s impossible to reverse engineer electronics to get at the smart bit, if it doesn’t reside in the product electronics.
Taking on the technology challenge
If the challenge of incorporating a vision system into a consumer product is taken on, what would it need to do? For every product developed, the operational boundaries and requirements specification need to be tightly defined. Plus there will always be cases where the system fails, and these situations need to be catered for as well. A next-generation robotic lawn mower would need to be able to see the edges of a lawn and navigate by building its own internal map, ideally without shredding toys, flip-flops, toes and hedgehogs. It should, in effect, be an affable electric sheep, and it would need to be engineered to be built at a price point suitable for the consumer market.
Seizing on the challenge of building an electric sheep, I pondered how it would see the world. Sure, we could give it the same vision we have, but the world is large, unstructured and hard to understand. So I started to dream of a sheep that could see just the grass and would ignore everything else. Such a technology would make the lawn boundary detection problem much easier and far less expensive to solve. After applying some real-world physics, I was able to build a camera system that filters the incoming light to allow a multi-spectral response, based on the fact that, like all green plant life, grass reflects infra-red light far more than the background earth or many other surfaces.
To breathe life into my electric sheep I also devised a robotic platform as a test bed for carrying the cameras and other sensors. This robotic device also incorporated vision-based navigation sensors, that again, had to meet the cost-nothing criterion. Again, the challenge was reducing the visual data to an amount that could be handled by a very low functioning processor or gate array. As a test bed, the system was sufficient, and a lot of progress was made on the fundamentals required to turn the ideas into something applicable to this industry sector. The challenge of leveraging a sheep-level artificial intelligence into a sub-dollar processor will need additional effort, I suspect.
Lawn cutting is an interesting area and a challenge for product development. In many countries, house owners cut their own lawns, in others the owners get other people to cut their lawns. In one country lawn mowers are a product for lawn cutting, in others lawn cutting is a service provided by a contractor. The shift from product offering to service offering is not only applicable to lawn mowers, but across many consumer sectors. The transition from consumer offering to consumer service offering to commercial offering is a challenge in itself, and the possibilities in these areas require deep and broad market understanding. Technology and imagination can rise to the challenge, but deep market insight is required to meet the challenge effectively and at an appropriate price point.
