American astrophysicist and cosmologist Neil deGrasse Tyson once commented: “In the early days of robots people said, ‘Oh, let’s build a robot’, and what’s the first thought? You make a robot look like a human and do human things. That’s so 1950s. We are so past that.”
It is also a view that is shared by the team at Helsinki-based ZenRobotics, which has launched a robotic waste separation technology entitled ZenRobotics Recycler (ZRR), designed to increase efficiency and lower the cost of waste separation.
“Unlike traditional recycling machinery that’s based on mechanic and electric components, ZRR is powered by artificial intelligence,” says Timo Taalas, CEO of ZenRobotics. “It is also the first commercially available robotic waste sorting system which offers one system for multiple tasks.”
However, before images from I, Robot or Wall-E threaten to distort the picture being conjured up of ZRR, think instead of a gripper on a production line and you will be closer to the reality. The clever part is the AI, which enables ZRR to sort different types of materials and objects of varying shapes and sizes, thereby reducing the need for the pre-processing of waste.
“Currently, our customers use ZRR for sorting metals, different grades of wood, various grades of minerals (gypsum, stone, concrete), rigid plastics and cardboard,” continues ZenRobotics’ CEO, who goes on to emphasise that ZRR “is not limited to these fractions alone. The system can also be trained to sort specific objects such as tubes and pipes or even plastic bags. A single robot arm can pick up to four different fractions with high precision, reaching up to 98% purity.”
At the heart, or should that be the ‘head’, of the ZRR system is the brain; software which analyses the data and controls the robotic arm(s). And here’s the clever bit: the ZRR brain allows the system to learn new fractions, enabling operators to react to changes in the waste system.
“Training can be used to introduce a completely new fraction, improve sorting quality of a current fraction or divide existing fractions into sub-fractions,” says Taalas.
“This kind of flexibility has not been made available in waste sorting before and it’s made possible by the significant research and development work done by ZenRobotics,” explains Dr Tuomas Lukka, a member of the founding ZenRobotics team – and, for anyone who remembers the popular children’s TV series Thunderbirds, is definitely the equivalent of Brains, having acquired his PhD at the tender age of 20.
“When I started this company in 2007, we had a succinct business plan to do something with robotics. There was a glass ceiling in the application of industrial robotics so we took a couple of years and went around various Finnish companies looking for problems that we could solve. We looked at different enterprises from bakers to makers of detergents, but it soon became apparent that one clear problem was grasping objects that were not the same shape. We decided to start solving that and recycling is a really good area for this application,” recalls Lukka. “If you send a robot to a car factory, it does the same thing 99% of the time, but recycling offers more complex challenges than manufacturing so we decided to make a robot that sorts waste for recycling. We are looking for more complex problems for robots to solve.”
ZenRobotics is already offering its second-generation ZRR2 which, as the name suggests, comes with two robotic arms.
So what do you get?
ZRR2 comprises a ZRR sensor unit, containing multiple advanced sensors; brain control software in a separate climatised control cabinet; robot arms and a smart gripper which can make up to 4,000 picks per hour per ZRR2 unit, and drop-off chutes for up to four chutes per arm. The third ZRR offering is the company’s Semi-Mobile, which also includes a 1.6m-wide sorting belt, drop-off chutes and internal electrification installed on a self-supporting semi-mobile platform.
Taalas again: “In manufacturing, the normal robot stops when it collides with an object, so we had to design ours to deal with these situations. It is also a modular plant so you can bolt an extra arm onto the plant.”
How does the system work?
• ZRR sensor unit scans the waste stream.
• ZenRobotics’ brain analyses data and controls the robots.
• ZRR identifies materials, objects and gripping points.
• The gripper picks the desired objects.
• Robot sorts multiple fractions in one spot.
“We train the robot as you would train a child until the robot continues to learn to handle different-shaped objects. It picks objects and we tell it whether it failed or whether it succeeded,” explains the CEO, before recalling with refreshing candour: “There has been a great deal of trial and error until we found a gripper that works. It can pick a 20kg object and estimates the weight of the object; if the object is heavy it approaches it more slowly, so there is intelligence there.”
So far the ZenRobotics team has focused on C&I and C&D waste. “These comprise all kinds of inert materials, e.g. concrete,” continues Taalas. “We tested it with scrap metals and the sorting of plastic bags. The robots are good at picking heavy objects. There is an area where it works best, e.g. bricks. However, if it comes to small objects such as pencils, there is no commercial value in picking up 4,000 of those.”
Teaching new tricks
Once a system has been installed, there is no need to call in the ZenRobotics team to teach the system new fractions.
“Customers can train the robot themselves,” states the CEO. “There is no code for different-shaped objects. You collect samples of the fractions you want to sort, e.g. a pile of wood or stones, and you feed the objects to the system, put the objects onto the conveyor and then you run the training script – and the AI learns the difference between metals and woods. It gives the customer additional flexibility. With changing legislation, you can upgrade the software. We had a case of a customer who was sorting wood, metal and stone, but the price of wood was so low, he wanted to separate A-grade and B-grade wood. We collected the samples and it took half a day to run the training.”
The system has different kinds of sensors, including a metal detector under the belt and near infrared (NIR) lasers that detect the height of objects.
“From those sensors you combine the data and megabyte of data that is gathered every second and a computer processes it,” continues Taalas. “We use the data to feed the AI and, based on these inputs, the robot can see the difference between the two grades of wood.”
With regard to costs, the price of one robotic system is €600,000-700,000. Pay-back can be within a year depending on the variables.
If a ZRR system is compared with a picking line comprising eight operatives working on two shifts, and depending on the materials, the ZRR system is said to achieve 2,000 picks per arm every hour, while one person per hour is estimated to make 800 picks.
“For a small operator where space constraints are an issue, you would be looking at two arms and eight chutes,” adds Taalas.
The CEO goes on to point out: “Power consumption is 20kW, while the Suez Helsinki plant is 50kW. We recommend clients to extend operating hours to get the most out of the system.”
Manufacturing of the system is carried out near Helsinki. Most of the parts are standard and can be sourced from mainstream manufacturers. This means the robots are manufactured with off-the-shelf components, but manufactured by ZenRobotics following its own design and assembly. The key technology has been patented.
Maintenance consists of a mere half an hour per day. “This includes greasing and dusting,” says Taalas. “The only wear parts are the steel gripper blades, which have been designed so you can unscrew a few bolts and replace the blades in a few moments.”
According to Dr Maciej Borkowski, ZenRobotics’ sales director: “It takes four months to deliver a unit.” He pauses before adding: “Basically, there is nothing scary about robotics in our operation.”
This means we can all rest easy at night knowing there is no danger of ZenRobotics’ machines trying to take over the world as we know it, but instead making waste separation simpler.
Fact file: Suez Helsinki
Employing the ZRR system, an unmanned robotic sorting plant at Suez Environment’s site outside Helsinki operates along the following lines:
• Input comprises mixed waste. “This reduces the cost of building in different conveyors,” comments Taalas.
• Screening and separation remove fines and 2D while 3D material is left for robotic sorting of four fractions into 10 sorting bins.
• Fines <100mm go to mechanical treatment while 2D material is destined for RDF or material recovery.
• Plant sorts multiple fractions simultaneously in one spot.
• Concrete, mixed inert, A-grade wood, metals, mixed rigid plastics, pipes and tubes and plastic bags are sorted by colour, or customers can train ZRR to pick new fractions themselves.
• Storage buffer of materials: ZenRobotics is checking noise output levels, particularly for urban areas. “We are measuring the noise levels, which currently stand at 85db. Next to the demonstration plant is a nature reserve, and the plant has been given permission to run 24/6 (not on a Sunday). In terms of dust, the robots are accustomed to a dusty environment,” says Taalas with a smile.