Robotic Waste Sorting: The Picks-Per-Minute Illusion

>Robotic Waste Sorting: The Picks-Per-Minute Illusion

Robotic Waste Sorting: The Picks-Per-Minute Illusion

2019-11-19T14:25:31+10:30 7th November 2019|

Recycling or waste sorting involves recognising precisely recyclable objects and extracting them as rapidly as possible to put them in the right bin.

Human pickers are fantastic waste sorting machines. Their eyes are colour and stereoscopic vision systems that can capture 24 images per second. Their brain can process all of these images, learn to recognize any target object, and calculate the trajectory of an object on a moving conveyor. Furthermore, the brain can develop a complex grappling strategy and command a person’s two arms and ten fingers to perform highly adapted and coordinated mechanical moves to optimize the extraction.

The problem with human pickers is that they are an increasingly scarce resource, expensive and inconsistent. They get tired, sick, distracted, intoxicated, bored, careless – you name it!

As a Materials Recovery Facility operator, if they show up for work, you will have to pay them no matter the fluctuations in their daily sorting performance and productivity.

So, what’s the solution to all of these operational challenges? Of course: A.I. powered recycling robots, which are inexpensive, fast, precise, reliable and consistent waste sorting machines.

As a MRF operator, that’s exactly what I need!” The remaining question is: “Can this robot really outperform my human pickers?” To answer this question we need to define sorting performance.

The Nature of Sorting Performance

In current and future recycling market conditions, quality of sorting is paramount; “Quality wins!“. What would be the point of sorting at light speed if the quality of sorting is poor and end-products are cross-contaminated with no one to buy them?

In the picture above, humans or A.I. would be hard-pressed to tell, without investigating further, which product is made of which resin. Only proven technologies such as Near-Infrared (NIR) or Hyperspectral (HYS) scanning can differentiate precisely these products. In the current market conditions if ‘Quality wins’, the primary quality of a sorting robot should be precision of sorting, not speed.

On the subject of sorting speed, what is the precision of a human picker? 70%? 95%? Your guess is as good as mine. It depends on the picker, the products being targeted, the time of day, the weather outside or if his team won yesterday…

Of course, we need these robots to be fast, but at what speed does the robot outperform the human? 35 picks-per-minute? 80 picks-per-min? The faster the better right? Wait a minute!

Human picks-per-minute vary greatly. You have guys that can use a scooping technique that can probably deliver up to 100 picks-per-minute! But, can they do this day-in-day-out? No. Somehow the industry appears to agree that sorting speed is  40 picks-per-minute average. Is this a true average sustainable day-in-day-out? Maybe. Nevertheless, let’s assume this as the benchmark to measure human sorting speed for light-weight items, not C&D material.

Robot Picking Performance

So, how many picks can your robot do? Well… in theory and according to the robot manufacturer’s specification sheet it can do 80 picks-per-minute! The work of two human pickers! Really? No.

It is not that simple. What the robot manufacturer’s specification sheet is telling you is that it can accomplish up to 80 picks-per-minute on an optimal trajectory, well within the robot work area. On a random set of trajectories moving to the outer edge of the work area, as necessary in waste sorting, the performance will go down significantly. Sure these robots can move crazy fast but not every move is as fast as the other.

The picking sequence has a greater impact on the number of picks-per-minute performed by the robot. On a 150 feet per minute (FPM) conveyor, the items will traverse the robot work area (4 feet) in 1.6 seconds. If several items show up at the same time it is unlikely they will all get picked up no matter the speed of the robot.

The important metric in picking efficiency is the robot work area which converts on a moving conveyor to contact time with the material. Just like humans, the more hands you have on deck the more sorting you will perform. No matter the picks-per-minute claimed by the robot supplier no one robot can do it all.

Slower multi-robots working in cooperation will deliver more Overall Equipment Effectiveness (OEE: www.oee.com) than the world’s fastest robot. Beware of multi-robots that do not really work in cooperation but that just divide the belt into two half belts and so do not significantly increase the work area or contact time. They surely increase the number of picks performed but they will never beat truly cooperating full belt robots that can pass ‘missed picks’ to the robots downstream.

Finally, grappling efficiency is also a very important metric to measure OEE. You can have the most sophisticated vision system, the fastest robot but if you can’t extract the product from the belt you are dead in the water. Keep in mind that in a dirty environment, ‘suction sucks’.

An educated sorting robot buyer will look beyond the picks-per-minute claims and try to truly measure the OEE of the contemplated robotic sorting system by looking at recognition technologies, robot speed, work area and grappling efficiency of the proposed system.

In a follow-up article in this series, we will go beyond Overall Equipment Effectiveness (OEE) and dive into the measure of Return on Investment (ROI) for a robotic sorting system. To be continued…