The commercial marijuana business could be worth more than $57 billion in the next decade. As the sector comes out from the shadows and rapidly scales up, enterprise services and suppliers are springing up faster than, well, weeds, and they're saturating the industry with new technology.
A company called Bloom Automation is on the verge of deploying smart robots that can harvest the sensitive crop at scale more efficiently than human workers.
Automation has become a major force across the agriculture industry. Farmers big and small are adopting everything from self-driving tractors and robotic harvesters to drones that monitor crop health and watering needs.
Automating marijuana harvesting has been a trickier proposition.
"It's been done by hand because the product you want is very specific," Jon Gowa, CEO of Bloom, told me on a recent call. "A traditional machine would chop it up. At the end of the day cultivators are selling this for quite a lot of money, so human harvesting has made sense."
But that's changing as increasing competition has created a glut in the market in places like California and Colorado, where marijuana is sold legally at the state level.
"You used to be able to sell a pound of marijuana for $3,000 two or three years ago," says Gowa. "Now, in Colorado, prices are closer to $1,200 or $1,000 per pound wholesale. If a supplier's processing rate is $150 per pound, they're going to run into trouble."
The business of marijuana is going to increasingly get attention because there may be stock market riches to be had. Marijuana may be the new Bitcoin.
For instance, Tilray, one of the early leaders in the pot industry, traded nearly $6.5 billion in shares on U.S. exchanges Sept. 19. Tilray closed the session up more than 40 percent, but was up more than 90 percent at one point. Other pot stocks such as Canopy Growth, Aurora Cannabis and Cronos Group had active trading sessions.
Compliance and cultivating costs, in addition to processing, make growing cannabis an expensive proposition. In California, there's now speculation marijuana may drop to a floor of $500 per pound wholesale.
Automation has traditionally been the way location-dependent suppliers in industries of various stripes have curtailed labor costs. But harvesting marijuana, which is grown in a greenhouse, presents unique challenges. Some larger suppliers are currently using tumblers to sift out the desirable parts of the hemp plant, but that can damage buds.
"We saw a need to automate the process in a way that treats it as an organic product as opposed to a homogenous product."
The robotic harvester uses machine vision and path planning algorithms to isolate clusters of flowers. The system incorporates a back-lit time of flight camera, which measures depth, as well as a machine vision camera.
"The system segments the plant into three parts, the flower, branch, and leaf," explains Gowa. "We use a conventional neural network and a supervised machine learning set."
The Bloom team has marked up more than 6000 images and counting by hand to teach the system to distinguish between the segments. The algorithm is now 97 percent accurate.
With the parts of the plant identified, the system can plan a path, directing the clippers to the flower.
Bloom estimates they'll soon achieve two-to-one efficiency over human harvesters. "Our estimated ROI for the average cultivator purchasing 6 robots to replace or augment 12 human trimmers is 8 months," says Gowa.
Cannabis is an industry in flux, but it looks increasingly likely the future of the sector will be autonomous.