Diving deep into robotics

Recycling Today Media Group’s MRF Operations Forum is an event specifically dedicated to plant operations, offering attendees an in-depth look at issues facing material recovery facility (MRF) operators. This year’s event takes place Oct. 22 in Chicago. 

Registration for the MRF Operations Forum is already open. Registration rates are designed to be cost-effective, with special discounts available for multiple attendees from the same company. Also, those who attend both the MRF Operations Forum and the Paper & Plastics Recycling Conference, which takes place at the same venue Oct. 23-25, benefit from even more discounts. 

At this year’s MRF Operations Forum, manufacturers and recyclers will explore how robotic technology works and how it can be applied at MRFs. In the opening session titled, “Robots in Action: How They Work from a Supplier’s View,” which takes place from 8:45-10:15 a.m. Oct. 22, manufacturers will share information on robotic mechanisms, physical integration and AI integration.

Recycling Today connected with speakers from that session to learn more about why robotics and artificial intelligence (AI) are critically important to MRF operators and what MRF Operations Forum attendees can expect to learn in the session. 

Recycling Today (RT): How far have AI and robotic technologies come in the last few years?

Thomas Brooks, director of technology development at Bulk Handling Systems (TB): The developments in AI and robotics continue to accelerate. Not only is the technology on a fast trajectory but at an ever-increasing rate. Where we see the greatest growth is in the grasping and machine vision. Both key areas make the implementation of AI and robotics globally effective and ultimately applicable. In robotics we see this in move rates and grasping technology. The research that is taking place in grasping or material manipulation is amazing. This is not only in the hardware but also in the application of AI to the process of grasping and maintenance.  

Eric Camirand, founder and CEO at Waste Robotics Inc. (EC): In the last five years, AI has become very important in multiple applications that we use in our daily lives. In that time frame, several algorithms (Google TensorFlow, Berkeley Caffe, Torch, etc.) were released in open source and helped fuel this very rapid development. Before AI, sorting through waste on a conveyor full of random objects was virtually impossible. Development of this technology requires very specific skills that are not common in the waste management industry. Only a handful of companies have developed these technologies and are bringing them to the recycling market.

Matanya Horowitz, founder and CEO at AMP Robotics (MH): In the last few years we’ve seen AI and robotics evolve from a proof of concept to a fully commercialized set of tools for MRFs across the world. When AMP deployed its first system in 2015, it was considered a minor miracle that the system could pick material at all. Since then, we’ve seen the technology evolve from low pick rates to 80-plus picks per minute, handling burden depths that are difficult for people, and deployed on a wide variety of lines within MRFs from quality control to fiber lines. What’s exciting about the technology is that it’s not only opening up new doors for MRFs, but the pace of technology is only accelerating. We’re seeing improvements in the technology on nearly a monthly basis.

David Marcouiller, executive vice president of sales engineering at Machinex (DM): [AI and robotics have come] from infancy to commercially viable in our sector. Today, it is very common to find robots and AI integration at all major MRF upgrades or new MRFs.

Rainer Rehn, chief commercial officer at ZenRobotics Ltd. (RR): In industrial automation, robots are generally used in a defined, structural environment. When a robot is used to produce a car, it knows what to do for the years to come. It knows what the objects look like, how they arrive on the production line and what it has to do with them. The challenge for a waste treatment robot is the unstructured environment. Nobody knows exactly what the waste will look like or what the composition will be. That’s the ultimate challenge for sorting robots. Thanks to AI, we can make the robots flexible and adaptive so that they can recognize, grab and sort objects in this unstructured and complex environment. That’s the main difference when compared to a traditional industrial robot. Industrial robots are components that can’t do much on their own and are adapted to specific requirements. ZenRobotics offers a highly integrated solution that’s optimized for the purpose of waste sorting. This means we don’t just have a robot, but we have an intelligent combination of image recognition, machine learning, motion control with firm gripping and throwing and many years of experience in waste processing. All this makes the solution special because it is optimized for a specific application. ZenRobotics applied AI and robots to a real waste-sorting environment in 2010. Building on the latest research in the field, the technology was designed to bring the efficiency of automation to the waste industry. Thanks to AI, industrial robots could now work in a new, more complex environment. 

RT: What are some misconceptions people have about robotics and AI? 

TB: This is a great question and the response ranges from “robotics and AI solve every problem” to “they are the bane of mankind (i.e., Terminator/Matrix).” One that we see quite often is the view that robotics and AI are the same. For some reason, most people see robotics as the incarnate manifestation of AI. This is not surprising, as society was most willing to accept AI-operating robots first. This holds true for the recycling industry also. All of the early adopters of AI in the recycling space focused on robotics. Robotics have been used in industrial automation since the late 1950s, [but] our industry has had limited exposure to them. This created some tough misconceptions at first about maintenance and support for robotics and machine vision systems. Since BHS also produces optical sorters, we were able to incorporate much of our expertise in the hardware and software for easing the transition to working with robotics. While BHS offers products that have the two closely integrated, like our AQC, we utilize AI for so much more. 

EC: AI and robotics are not perfect. Just like humans, they have limitations. AI works well only when trained with big data. Google and Facebook collected a phenomenal amount of data more specifically tagged images of our daily lives (dogs, cats, people, cars, etc.) that constituted the big data required to train very impressive AI recognition applications that are now coming online. However, there are very few images available of truck-compressed dirty recyclables. Capturing and tagging these images is a challenge. Moreover, the packaging is constantly changing. So if you train your staff or your AI to put the “Tide” detergent bottle in the HDPE (high-density polyethylene) bin and all of a sudden the company starts packaging their detergent in a PET (polyethylene terephthalate) bottle you will have your staff or your AI robot sorting wrongly and cross-contaminating plastics and degrade its value. At Waste Robotics we have recognized this issue and use beyond-visible AI to ensure we can readily deliver quality of sorting.

MH: I think people believe the technology is more nascent than it really is. At this point, we’ve had over 10,000 hours of operation on some of our robots. These systems have become truly reliable, low-maintenance pieces of equipment. It’s not a curiosity anymore; it is a core part of many of our customers’ operations.

Photo by Austin Friedline with Phierce Photography

DM: People still think that across the board, on a given quality control position, humans are better. At the end of the day, a robot-staffed position does a better pick rate both in terms of efficiency and product purity. People still think that robots are unproven. There are many success stories in the field now proving their viability in our sector.

RR: Nine years ago, robotic waste sorting was an exotic technology. We had to put in a lot of effort and pioneering work to educate the market. AI was seen as something that’s too difficult to grasp. Today, as AI is incorporated in our daily lives, through search engines and mobile phones, the concept isn’t regarded as intimidating. Although, as the understanding widens, we like to emphasize that AI is only a tool for improving the performance and an integrated part of our solution. Our customers do not have to be experts in AI to operate the units. Details about the best AI algorithms, neural networks or learning systems, are not the key. The only thing that actually matters is the true real-life performance of the machine as a whole. There is little use of a multilayered neural network if the gripping is not successful.

RT: How can robotics help MRFs improve their operations?

TB: I came from a completely different industry which focused on Department of Defense and biomedical applications. In these spaces, the boundary conditions and variables to solve problems are very well defined. Coming into the recovery and recycling industry was eye-opening. This may be the toughest industry in the world to define. This is driven by variation and interconnectivity in every aspect of the business: inbound material, sociologic shifts, commodity prices, government regulation, workforce availability, etc. As we automate aspects of the sort line and facility operation, this gives MRF operators the bandwidth to focus on expanding ways of growing revenue and increasing profitability. One example of this is shift times. Current operational schedules are driven by human need: start times, lunch, breaks, stop time. With a fully automated system, the operator has complete control over this and can run the operation in a way that is most profitable.     

EC: Predictability, precision, constancy, speed, reconfigurable sorting is what robots offer to MRFs. Human labor is inefficient, expensive and unpredictable. In order to sell sorted end products, quality wins. Robots can deliver quality. Robots will significantly help MRFs when they decide to operate differently. A robot at the end of a human sorting line will only work as many hours as the other humans on the line. Robots should not sit idle or sleep. To be efficient, robots need to be deployed in 24/7 sorting applications. Fully automated MRFs will not look the same as the MRFs we have today. 

MH: Robotics can help build consistency into a MRF’s operations. The robots don’t mind working in a wide variety of conditions and never lose their focus. Additionally, we’re finding that facilities get the best use out of their system when they have the system pick as many materials as possible. This ties nicely into what MRFs are looking for: to extract as much value from the material stream as possible, which of course requires as much material to be extracted as possible.

DM: [Robotics can help MRFs] with sorting efficiencies and reducing labor rates, thus reducing overall operational expenses. The data gathered by the AI of the sorting robots can now supply the operator with data that was not available from the sorters before. AI and robotics are important to help with the ever-increasing labor costs, manpower and sorters available with current unemployment rates and to help with the cost of turnover.

RR: The main benefit is increased efficiency and productivity that automation allows. You can run the plant nearly 24/7 with constant speed, as robots work continuously without stops. Second, the waste industry needs more sophisticated technologies for improving the purity of end fractions. Sensors and AI software allow more versatile sorting capabilities. For example, the robot can be trained to sort specific objects, not only materials. This gives businesses increased flexibility and opportunity to develop new business opportunities and provide high-quality raw materials that can be marketed locally. Finally, AI and digitalization also allow more data about the waste, which helps companies to improve and monitor their operations. Optimizing requires data.