1. The Economics of Robotic Recycling
Despite the fact that an estimated 35% of Americans recycled upwards of 70 million tons of waste in recent years, the $6.6B US recycling market continues to be on the verge of crisis for one simple, but profoundly important reason: it is extremely difficult to recycle the vast majority of materials in a cost-effective manner. Unfortunately, it is often cheaper for companies to use “virgin” input materials in their manufacturing processes than recycled ones.
The moment that your recycling is picked up from the curb, it begins a complex and labor-intensive journey through one of America’s 300 materials recovery facilities (“MRF”), where single streams of waste pass through a maze of sorters and sifters, conveyers and chutes, and pickers and packers to isolate like-recyclables. These facilities are staffed with a large manual workforce, which, until recently, has been the only way to sort difficult materials such as plastics…
…Enter the sorting robot. Over the last 5 - 10 years, significant progress has been made in the accuracy of computer vision systems that can categorize recyclable waste and in the speed with which robotic systems can articulate and ‘pick’ items from a conveyer belt. These innovations have breathed new life into the recycling industry. Robotic sorters are a promising solution to cost and quality issues that have plagued the industry for years. Below I will discuss current challenges in the recycling industry, robotic sorting tech and its potential cost savings, and emerging startups in the field.
The Financial Challenges of Recycling
Cost Competitiveness: Recyclers have had trouble being price competitive in certain recyclable categories. For example, in 2017, the estimated average cost to produce virgin (non-recycled) PET (polyethylene terephthalate) plastic was $0.52–0.56 per pound, while the cost to process and produce recycled PET was estimated at $0.60–0.65 per pound. Virgin PET is a crude byproduct and given that the price/barrel of crude today (~$52) is consistent with 2017 levels, these pricing pressures are likely still relevant.
Contamination: The average US contamination rates among communities and businesses sits at around 25%, which means that roughly 1 in 4 items placed in a recycling container is actually not recyclable through curbside programs. Contamination is a major problem because it increases the cost of processing recyclables (makes sorting more difficult) and it poses a threat to the quality of the recyclables.
China: Prior to 2018, roughly 40% of US paper, plastics, and other recyclables had been shipped and sold to China. However, on 18 July 2017, China notified the World Trade Organization that it would be imposing a ban on imports of certain kinds of solid waste by the end of 2017, citing domestic environmental and public health concerns associated with contaminated foreign waste. US recyclers once turned a small profit by sending low-value plastic waste (known as “MRF residuals”) to China for processing (cheap labor); however, this is no longer an option and MRF residual waste is now, more often than not, either incinerated or sent to a landfill.
The Human Challenges of Recycling
Workplace Safety: According to the Bureau of Labor Statistics, the recycling industry had the sixth-highest fatality rate among all US civilian jobs in 2019. The fatality rate was 35.2 deaths per 100,000 full-time workers. In particular, MRF sorting line workers are at a distinct risk due to their direct contact with waste materials, which can sometimes include sharp objects (e.g. syringes), toxins, or explosives.
Employee Retention: A combination of factors including workplace safety, the monotony of working tasks, lack of career trajectory, and low wages (the Bureau of Labor Statistics estimates that recycling sorters at MRFs received a median annual wage of ~$28K in 2019) makes recycling a challenging hiring market. In 2019, Waste Management, the largest US recycling operator (95 MRFs, or roughly 33% of the US market) reported a relatively high annual employee turnover rate of ~22% in 2019.
Productivity/Accuracy: the best estimated average productivity of a human picker is roughly 40 picks-per-minute. A worker’s pick accuracy is also likely to decrease as the workday progresses, which can lead to contamination issues. Furthermore, workers can only work a finite number of hours per day, and if a MRF wants to operate 24/7, shift changes (and inevitable downtime) will be required.
No Economies of Scale: the scatter plot below charts the ratio of MRF waste sorters in a given facility to its total number of employees. It is notable that the ratio remains relatively constant (50%-60%) regardless of facility size, implying that there are no labor economies of scale in recycling.
What is Robotic Recycling and What Does it Solve?
Robotic recycling technology can be best characterized as the combination of AI/computer vision software and the physical manipulation of robotic systems to sort waste faster and more accurately than a human sorter. The technology’s 4 core functions are detailed below:
- Identify: effectively all solutions use a combination of computer vision and machine learning/AI software to properly detect, classify, and track recyclables within waste streams as they pass under an optical reader. Most systems claim a classification accuracy of >95%.
- Pick: After recyclable classification, a robotic system (typically mounted overhead of the waste stream) will grab or ‘pick’ a single recyclable. Smaller and faster suction methods are typically used for lighter recyclables (plastics, glass, etc.); whereas larger, claw-like robots are employed for heavier items (metals, wood, etc.).
- Sort: Once a recyclable is picked, it then is sorted and deposited into a receptacle of like-recyclables. Maintaining the purity/quality of sorted bins is critical to maximizing the resale value of recyclables.
- Monitor: the use of computer vision in identifying recyclables within the waste streams facilitates the collection of a significant amount of data that can be used to inform quality audits and fine-tune operational decisions.
These systems aim to remove humans from the waste sorting equation, which will, in turn, reduce workplace injuries and fatalities and will reduce the overall labor burden that an MRF faces. Naturally, the reduction in sorter headcount will take cost out of the sorting process (assuming the cost of the robotic system is less over time) and improve the unit economics of recycling. Furthermore, the increased pick quality/accuracy will reduce the number of recyclable bales that are scrapped - further improving profits.
If you are a more visual learner, a video of one of AMP Robotic’s sorting robots is linked here.
Let’s do the Savings Math…
My calculations below suggest that, over time, deploying and utilizing robotic sorters is 4x-6x cheaper than employing human pickers due to a combination of pick speed and pick accuracy improvements:
It is notable that the calculations only address variable labor costs and do not comprehensively quantify the hidden financial impact of contaminated bales of recyclables. Contaminated bales are more prevalent with human operators (lower pick accuracy). When a certain percentage of a bale is contaminated (typically greater than ~5%), the entire bale is usually considered unusable and is thrown away.
5 Startups to Watch
As of early 2021, AMP Robotics has generated a lot of buzz around robotic recycling solutions after announcing its $55M Series B led by XN (NY-Based hedge fund) and involving Sequoia Capital, Google Ventures, and Valor Equity Partners, among others. According to Forbes, the Colorado-based company has sold or leased 100 of its AI-powered robots since 2017 to more than 40 recycling facilities in North America, Europe, and Japan. While the cost to buy one of AMP’s Robots outright is $300,000, the company also offers them through a $6,000/month lease, and it is estimated that AMP achieved $10M in revenue in 2019. The company is led by Matanya Horowitz, a CalTech Robotics PhD, who saw an opportunity to improve the unit economics of recycling with high-speed robotics. AMP’s robot works at an industry-leading rate of 80 picks per minute (~4,800/hour). A great interview with Horowitz can be found here on ClimateTech VC’s blog.
Greyparrot is a developer of waste recognition software intended to monitor, audit, and sort waste at scale. The company’s AI-powered computer vision can be used to determine purity levels of bales of recycled materials, identify hazardous materials or contaminants, and inform operational quality audits and specifications through live dashboarding/data analytics. The company claims that the software can identify 100% of waste flows and can be integrated/embedded into a range of smart systems and hardware including smart waste bins, smart trucks, and sorting robotics. The company was named in Business Insider as one of the “25 most innovative startups tackling climate change in Europe” and touts the former CTO of the UK’s Royal Mail, Jochen Alt, as its CTO.
Unlike its competitors above, Ishtiva Robotic Systems (“IRS”) takes a two-pronged approach to waste sorting that combines overhead suction-based robotic sorting (similar to AMP’s) with compressed air-powered sorting. The latter process uses high-powered blasts of compressed air to isolate and eject HDPE plastics from a single stream of mixed waste. Both processes are powered by proprietary AI/Computer vision software that the company claims is 95% accurate in waste identification. The company claims that its combined system can achieve 3,000 picks per hour.
Waste Robotics is developing a variety of overhead ‘grappling’ robots capable of sorting organic recycling, single-stream recycling, and construction/demolition recycling. Unlike the suction-based robots of competitors like AMP and IRS, Waste Robotics’ grapple attachments are capable of sorting heavier materials like bricks, metals, and wood; however, the requisite clamping action associated with sorting heavier items inhibits the robot’s pick speed. In one installed setting, a Waste Robotics sorter was capable of processing about 15 tons of material an hour, at 30 picks a minute (1,800/hour), and with upward of 80% target material identification and separation. The company uses propriety AI/Computer vision software to sort waste and has partnered with FANUC (Japanese developer of industrial robots) to accelerate product development.
ZenRobotics, the most tenured company in this set, has been working since 2007 on two unique solutions to robotic waste sorting: a “heavy picker” and a “fast picker”. The heavy picker is capable of sorting objects up to ~66lbs (30kgs), and a single arm can pick at a rate of 2,000 objects per hour. This ‘grappling’ solution is most effective in applications of construction waste, scrap metal, and stone/inert waste. The fast picker, on the other hand, can pick at a rate of 4,000 objects per hour and uses a nimble, suction-based apparatus for sorting. This solution is most applicable to plastic, foil, paper, and other lightweight waste streams. ZenRobotics runs off a proprietary AI/computer vision software, called ZENBRAIN.
Outlook
At the market level, it is clear that the US recycling industry faces international trade headwinds; however, domestically, the newly elected Biden administration and democratic legislature have pledged to reinvigorate government-backed efforts to combat climate change. Biden’s proposed day-one executive orders address an ambitious array of climate-focused initiatives, from limits on methane pollution to ensuring every federal infrastructure investment reduces climate pollution. Some analysts suspect that oil prices will increase under Biden, which could stimulate demand for recycled plastics as the price of virgin plastic (again, a crude byproduct) increases.
At the technology level, computer vision has reached maturity and far exceeds human capabilities in terms of image classification (estimated ~5% error rate in humans vs. the <0.5% error rate of AI). Furthermore, as startups further refine the actuation speed of their robotic systems, the ROI of robotic systems will become undeniable. In the US market, approximately 95 percent of the throughput in today’s U.S. MRF environment is processed using equipment sold by just five producers, with the top two makers having a combined 63 percent of the market. This stat implies a significant lack of competition (and likely a lack of innovation) amongst incumbents. That said, I believe that there is significant white-space for more nimble startups to gain traction in this market.
Investors have bet on AMP Robotics as the early leader; however, I don’t think the buck will stop there - there will be room for multiple winners as a growing global population exacerbates the war on waste.
Note: opinions in this blog are my own and not the views of my employer
