AI-Powered Waste Sorting Robots from Sadako
It’s always those industries that don’t sound very sexy where you can make some real money. Take for example the waste management industry. Nobody likes to deal with garbage, but someone has to. You can’t just dig a big hole and fill it with garbage like they do in some places. “That’s bad for the planet man, so what you need to do is sort every single piece of that isht into a pile that contains a certain type of material so it can be recycled or even burned for energy” said the State of California to their 478 municipalities. (Actually, burning for energy isn’t really economically viable because it creates such an environmental mess.) This means that we actually have massive plants that are constantly processing our garbage using technology that is more advanced than you might think.
In order to get our hands dirty, we sat down with the Founder and CEO of Sadako Technologies, Eugenio Garnica, along with the CFO, Belen Garnica, to get some of the latest dirt on the waste management industry.
Founded in 2012, Barcelona startup Sadako Technologies sits on the outskirts of the city in a nondescript, but lovely neighborhood where their employees do nothing all day except teach deep learning algorithms how to pick up garbage and take siestas. With about $1.7 million in total funding (equity and debt) and another round expected this year, the company’s value proposition is using robots to sort garbage, but in order to understand that better, we need to first look at how things happen today.
The way garbage is processed today is surprisingly sophisticated, even though the technology is about two decades old. Your typical garbage plant will use various types of machines to sort Municipal Solid Waste which we will henceforth refer to using the acronym MSW. All this MSW then goes from the garbage trucks into a Material Recovery Facility (MRF) and out comes money. Well, not really. There are two main revenue streams for an MRF; payments per ton from the municipality, and the dollars from recycled stuff (it’s probably like a 80-20 or 70-30 breakdown with recycling being the lesser of the ratio). A typical MRF can handle 70 tons of waste per hour, which means one facility can handle a population of one million garbage generators.
Here’s a quick summary of how this happens. The basic machines “tumble” the waste to separate sizes and types while the more sophisticated machines do stuff like “in-flight sorting” which is even cooler than it sounds. As the trash falls off the conveyor belt – in mid-air – they use infra-red to detect the material composition of the object and then shoot a beam of air at it to redirect the object – in mid-air – onto the appropriate path. The limitations of the present method (which they call optical sorting) are really twofold. Firstly, the infra-red sensors cannot tell us anything useful about the shape of an object. For example, plastic bottles and plastic trays have a meaningful difference in value that the present methods find impossible to sort. Secondly, they need to have an entire team of humans doing QA afterwards because the system is not so accurate (not to mention the humans that are also involved for pre-sorting).
We thought that the present garbage sorting technology was remarkable enough, but what Sadako is doing takes it to the next level. They first use deep learning to scan the garbage visually (using industrial cameras like those from Basler), and identify objects, while a fast-moving robot arm reaches down and grabs things off the conveyor belt faster than a Mexican on meth. As time goes on, the algorithms get better at identifying trash. They also do it as accurately as the standard MRF hardware which they compliment as opposed to replace. This also means that eventually, we won’t need to have a bunch of humans injuring themselves as they grab hypodermic needles and sometimes even snakes. This AI-powered garbage sorting robot is already being sold to MRFs thanks in part to Sadako’s commercial partnership with a U.S. company called Bulk Handling Systems (BHS) which builds those cool garbage sorting machines. In other words, MRFs use BHS machines to processes lots of MSW. Now, they do it better with Sadako’s AI.
Several years ago, Sadako started out by building a robot suction cup. The idea was to use the robots to grab things off the line that would be too expensive to have a human do today. In other words, they wanted to create additional revenue streams for waste processing facilities. With a proof of concept in hand, they approached BHS and proposed a partnership. Sadako decided to focus on the deep learning part of the solution and BHS would build the robotic hardware and enclosure. The end result was a waste sorting robot from BHS called Max-AI™ which uses Sadako technology and is consistently able to sort garbage as good or better than humans do now and it doesn’t take smoke breaks:
That machine you see above, Max-AI™, was their first install of Sadako AI technology in April 2017 to Athens Waste Management near Los Angeles. The second install was a robot with even more complex detection which made its way into a Pennsylvania waste facility, and they’ve also installed another one in the U.S. but they can’t say where. One is also headed to the U.K. and a few more to several other unnamed destinations. As for how the humans reacted to the robot, most did the sensible thing and started taking selfies with it.
In Sadako’s Barcelona office, we checked out a stack of Nvidia GPUs that they had training their deep learning algorithms. The sorts of things they’re solving for are particular objects like polyethylene terephthalate (PET) bottles along with other thermoset plastics that can be recycled. The bottles can either be full of air or crinkled, and also partially obscured by other bits of rubbish. The algorithm is able to handle all these cases in almost real-time with a very high degree of accuracy thanks to the powers of deep learning:
That ability to detect objects is also finding some other uses as well. Sadako was proud to tell us about their $1.2 million grant from the European Union that focuses on their ability to provide data from the garbage that’s passing through all these waste management facilities:
You can’t do that sort of thing with infra-red because it’s too expensive and the data is not rich enough. The Sadako computer vision hardware without the robotics is about the size of a large toolbox, and they could easily see an MRF having ten of them installed across the facility in order to learn all about the waste being processed.
When it comes to competition, there aren’t as many players as one would expect given how potentially lucrative this space is. Some estimates put the global waste management industry at around $500 billion with that number only expected to grow alongside global population growth. Sadako’s leadership team noted two companies to watch that are also sorting garbage with robots.
Founded in 2007, Finnish company Zen Robotics has taken in almost $16 million to build AI-powered sorting robots that are different in a number of ways, most notably being that they work primarily with construction and demolition waste. Oftentimes, construction companies will be required by regulation to sort the construction waste into neat little piles of concrete, metal, and wood. The metal is also worth some money. The Zen Robotics solution was a first mover in the robotic-trash space, though sensor-heavy, with the main use of artificial intelligence being used for gripping approaches as opposed to identifying the trash.
The latest Zen Robotics blog post talks about their “first move into MSW” as they offer a solution for sorting different colored bags of sorted waste which can be seen in this video. Even if they did turn out to be a more direct competitor, Sadako thinks there’s room for everyone. The below chart shows the current distribution of waste sorting robots across the United States:
The above chart also records the location of some more waste sorting robots from another competitor that Sadako mentioned, AMP Robotics out of Boulder Colorado. (The two companies actually met in a garbage sorting contest).
Founded in 2015, Colorado startup AMP Robotics has taken in about $250,000 in funding so far to develop their garbage sorting robot called “AMP Cortex”. They offer the machine at a fixed rate for sorting stations which lowers labor needs which are a MRF’s second-biggest operational expense after transportation. Here’s a short spiel on the basic benefits of these types of garbage sorting robots:
AMP’s products are designed to meet the demanding needs of Material Recovery Facilities: higher throughput, increase commodity revenue, better bale quality, and a fixed labor rate over time.
There’s an excellent article by Dan Leif over at Resource Recycling that goes into greater detail on why you’d want to use machines like these. Apparently, one of the hardest problems that waste sorting facilities have is trying to find people who want to work those jobs. A quote from the article spells this out pretty clearly:
The MRF’s robot, from AMP Robotics, is replacing the 1.5 workers formerly needed to pull cartons off the line, and those employees have been moved to other sortation spots. “This will help,” Keegan said of the AMP technology. “It will replace labor we haven’t been able to get.”
AMP is using ABB robots and claims a 1-year payback period. They’re also using deep learning to identify types of garbage.
Update 11/14/19: AMP Robotics has raised $16 million in Series A funding to bolster its operations in the nearly $12.26 billion waste sorting robotics market and drive product R&D. This brings the company’s total funding to $19.5 million to date.
The holy grail for Sadako is zero people touching the waste. As the CFO said, “let’s get rid of the jobs we wouldn’t want our children doing”. The CEO refers to these jobs as “the three D’s”, dull, dirty, and dangerous. There’s a whole “green feel” to what they’re trying to do, and we couldn’t decide if they should have been classified as an AI company or a robotics company. We ended up classifying them as a green technology firm because what they’re doing will help us recycle more than we would before and that’s good for everyone.
Pure-play disruptive tech stocks are not only hard to find, but investing in them is risky business. That's why we created “The Nanalyze Disruptive Tech Portfolio Report,” which lists 20 disruptive tech stocks we love so much we’ve invested in them ourselves. Find out which tech stocks we love, like, and avoid in this special report, now available for all Nanalyze Premium annual subscribers.