Why Man & Machine Can Work in Harmony When it Comes to Sorting Waste : IN-DEPTH: Booming Bots Taking Over at a Material Recycling Facility Near You

Just five years ago the discussion was very much on the ‘why’ and what robotics and Artificial Intelligence (AI) can do for industry, which has for decades relied on manual sorting. Fast forward to the present day and the industry has witnessed a huge uptick in the number of companies now engaged in this space.

Combining AI with robotic sorting to help pick recyclable items more quickly and efficiently, the discussion has moved on from the ‘why’ to the ‘how’, from the ‘if’ to the ‘when’.

Indeed, bots are starting to boom. Figures from Goldstein Research forecast that the market size for waste-picking robots could reach US$12.24 billion by 2024.

Testament to this, at the end of 2019 US-based AMP Robotics announced that it had raised $16 million in Series A funding, led by Sequoia Capital. This followed an installation of 14 AI-guided robots at Single Stream Recyclers in Florida and nine other locations, all processing plastics, fibre, metals and other materials.

By using advanced computer vision and machine learning, the company says its Cortex robotics system and Neuron AI platform “continuously trains itself by processing millions of material images and adapting to changes in a facility’s material stream”

One notable development WMW covered in the past was the collaboration between Spanish company Sadako Technologies and US firm Bulk Handling Systems (BHS). Combining the AI developed in Spain with the robotic arms from the US, the cross-Atlantic marriage led to the birth of Max-AI.

Since then there have been various developments. BHS launched the Max-AI AQC-C system that is comprised of a Max-AI Visual Identification System (VIS) and at least one collaborative robot, otherwise known as a CoBot.

Whereas many early waste-sorting robots were designed to work in isolation from humans, the CoBots can instead work alongside people.

Forget the stiff and mechanised-type robotic sorting actions of the past; these CoBots instead have padded arms that move in a more fluid motion. Less robotic and stiff; more elegant, fluid and, if anything, more human.

For film buffs, imagine the arms of Spiderman’s archnemesis Dr Octavius crossed with the huggable Baymax inflatable robot from Disney’s Big Hero 6.

Thomas Brooks, director of technology and product development at BHS, says this is part of developing what he calls “collaborative solutions”.

“We started to realise that most of these facilities are actually designed for people,” he says, adding: “So we looked at what options do we have to develop collaborative robotics, while adding to our original AQC product line.”

Brooks says the AQC-C was borne out of a need to apply robotics to supplement the growing trend of labour constraints and fitting in sorting solutions to environments “that aren’t conducive to applying a robot”.

“With system retrofits, the structure that is required to support and safeguard larger robotic sorters is often too large or heavy to work without significant modifications,” he says. “Alternatively, a CoBot can be installed where a person works in just a few hours.”

In short: humanising robotic sorting to fit alongside traditional waste pickers.

The CoBots have also learned a new trick: they can throw.

Brooks adds: “When we think about what humans do really well, people do a great job of detecting things and, using our hands, we are able to move in different and unique ways to pick the object up.

“What we’ve done with the CoBot system is to start to leverage some of that. The grasping has been improved and we now have the capability, similar to the way a person does, to toss material across the belt.”

Watching a video of the CoBots in action, the simple throwing or tossing action humans take for granted is impressive when witnessing the same dexterity for the grab and throwing – with accuracy – from a robot.

In one scene, two arms work simultaneously: one grabbing a plastic bottle, the other an aluminium can, but separately sorting and throwing the objects over the conveyor belt into separate containers. With robotic precision, inevitably they hit their targets.

The only thing lacking from a comparison to humans would be the robotic equivalent of a high five.

Brooks recently hosted a site tour following a plastics recycling conference in Nashville. For a demonstration, BHS linked two robots together so they were “throwing to opposite sides of the belt”.

Up to four robotic sorters can be added behind each Max-VIS system, the company says, with each sorter able to sort up to 40 picks per minute on up to three different material types.

The new Max-AI product line includes VIS in its standard design with all equipment rather than being incorporated into the equipment structure. This enables VIS to be installed independent of a robotic or optical sorter and this distinction benefits Max-AI customers in several ways.

The neural network AI is trained for each installation, which takes place after VIS gathers data from the material stream. When VIS is installed beforehand, the robotic or optical sorters can be installed when the AI is trained and optimised.

“We can now talk not just about collaboration between people but also collaboration with robots working with other robots,” adds Brooks enthusiastically.

Canadian company Machinex’s system, called SamurAI, was first installed at the Lakeshore Recycling Systems (LRS) facility in Chicago.

Located on the containers line, after an optical sorter ejecting PET and a ferrous magnet, the robot picks three types of products: coloured HDPE, natural HDPE and asceptics.

There are now 12 installations in total across the US and Canada, with discussions taking place in France.

As well as the “60 to 70 manipulations per minute”, Machinex says its USP is the “power of its gripping tool”.

“We have a really powerful system but the beauty in where we are right now is there’s no more filter: everything goes through the hose,” explains Jonathan Ménard, executive vice-president sales & strategic positioning.

The company aims to increase the coverage of its robots through a new leasing programme. Similar to when you lease a car, payment terms are available in 36, 48 and 60 month terms. According to Ménard, this switches the budgetary decision on taking on a SamurAI from capital upfront costs (CAPEX) to operational (OPEX).

So it’s essentially the ‘Netflix-type’ model for waste-sorting robots?

“It’s almost like an Amazon-type model where it’s a leasing programme, but it’s an all-inclusive leasing programme,” he says. “This includes the spare parts, the wear parts as well, and the subscription to what we call the Mach Cloud, in which we make sure we always have the latest in neural networks so your AI is always at the optimum.”

In short, Machinex takes the robot and integration costs and bundles them together to deliver a turnkey system, which customers pay for on a monthly basis.

Similar to car finance packages, at the end of the leasing period waste managers can either hand back the robot or pay the remaining finance to take full ownership.

“We felt this will probably unlock a bigger turnover for these robots going in that direction as opposed to pure sales,” he says, adding that such a model could help to unlock the European market.

Across the Atlantic, Finnish company ZenRobotics – perhaps one of the earliest household names in this market – has been continuing its global rollout, despite the increased competition from the US and Canada.

The company has installed its traditional heavy-picker robots for construction and demolition (C&D) waste in Australia, China, Singapore, Japan, Europe and North America.

In Europe, Sweden and Switzerland have been big markets for the company and with a large project being finalised in Finland, its home country could be home to the most robotic systems globally.

For lightweight material such as packaging and dry mixed recyclables, ZenRobotics has launched its “Fast Picker”, which is also powered by the company’s AI system, “ZenbrAIn”, also used in the heavy picker.

The AI in the latest Fast Pickers collects data via an RGB sensor, which Juha Mieskonen, head of sales at ZenRobotics, says is “more similar to humans”.

Estimated to deliver 4000 picks per hour (66.6 per minute), the company says the ability to readily update the system for “performance-tuning or new fractions” has not been available in waste sorting before.

Currently the heavy picker makes up around 70% of company revenues and the fast picker the remaining 30%. However, the company eventually wishes to get this to a 50/50 split.

Mieskonen welcomes the increasing competition in the robotics market for waste.

“For us, the way we see it is that it’s just confirming that we have been on the right road,” he says. “So it’s confirming that there is need for this technology and maybe we were a little bit ahead of our time.”

He adds that the market is no longer asking what robots can do, or why they should have robots. Instead, they’re refining their questions as customer awareness has grown.

Repeat orders are also helping the Finnish company to grow.

Malmö-based waste management company Carl F was the first company in Sweden to apply robotic waste sorting in 2017. The company added a third robot arm to up the production and further increase the sorting rate.

The site was sorting metals, wood and inert fractions from mixed C&D and C&I waste. The third arm will enable an increase in throughput to simultaneously sort out plastics as well.

Previous results from the adoption of robots have included reducing the level of material sent for incineration from 75% to 40%.

Mirroring manufacturing to enable human-less MRFs

Juha Mieskonen believes that the increase in robotics will enable waste sorting to eventually become like other manufacturing, such as automotive engineering.

“Ultimately, our vision has always been that we would convert and have the waste industry just mirror manufacturing,” he says. “In manufacturing everything’s very fine-tuned and automation enables everything to work really efficiently.”

He adds that “data has a big role to play” to help really know “waste in great detail which will then help to also recycle in a much higher degree of purity”.

“This is one of the pieces that we – and I know others – are working on,” says the head of sales. “The human-less MRF is something that’s almost reality or even reality nowadays.”

Machinex’s Ménard believes the industry is getting closer to the vision of the “MRF 4.0”.

“It’s a global vision, right? It’s not just about the machinery but the complete systems and the interconnectivity between the machines,” he says.

He gives the example of when there are too many PET bottles, you can send the signals to the optical sorters to be more aggressive and the robots to be more active. “The sooner you know the composition of the product, the better it is,” he adds.

“So instead of being reactive at the end of pipe, if we can actually monitor the composition early on and have a feel of how fibrous rich the material is, or glass rich, or container rich, we can actually automatically adjust the conveyor speed, screening and angle adjustment. Like I say, optical aggressivity.”

Brooks from BHS says as well as the CoBots, equal efforts are going into integrating AI into optical sorting systems.

“Collaborative robotics is such a unique space,” he says. “If we look at the entire landscape of an MRF, we can start to think about it from a robotic standpoint and really optimise it as a whole.

“This is one area where I’m continually pushing our team, making sure they're interfacing not only the robotics and the AI but all of these other pieces of technology as well.”

The director of technology and product development adds that real-time material composition is where this is headed. Facilities that have all of those technologies feed information back to a central database hub. Users can then pull out data and analyse it in a number of unique ways.

“The next step is, how do we start to make decisions on that data in a way that the facility runs with different machine configurations?” he
asks. “That the sub-system communicates within itself to basically start making decisions.”

The term “AI” can sometimes be overused for PR or marketing advantages.

Yet what Brooks describes definitely falls under AI: intelligent decisions being made by machines, without human input, to improve operations autonomously.

With MRFs having to improve the quality of their outputs as overseas markets close the doors and domestic ones become more demanding, this development could be a welcome addition.

However, the very phrase “human-less MRF” doesn’t really do the developments justice and could strike fear into workforces reliant upon MRFs for work. We don’t talk about “human-less” car manufacturing – it’s accepted as the way in which the industry evolved using robotics.

Recent developments highlight that man and machine can in fact work in harmony. It’s not a case of human pickers and that “robot in the corner”. In the future, both will work on picking lines, without a fear of injury from wailing[BZ1] , mechanised arms or AI-counterparts taking jobs.

The robotic sorting market is clearly heating up. Added competition on both sides of the Atlantic will drive companies to continually improve, refine and make better systems. Without competition, companies can become complacent.

And in the race to process, sort and recover value from increasing populations, there is no room for complacency. Robots are here to stay and will likely soon be sorting in an MRF near you.