Editor's note
This week's sharpest tool sits in a nursery.
There is plenty in this issue that could headline a week on its own, from drones to supercomputers to a harvester maker reinventing itself as a planter. All of it is worth your time.
But the standout this week is the quietest of them. SCA put a vision model on the line at Bogrundet to judge planting cassettes in about a second each. It is not dramatic. It is just extremely clever, pointed at a task that is repetitive, monotonous, and done millions of times a season.
I think that is where a lot of the next wave lands. Not the robots we film, but the dull, monotonous work we have always just absorbed. I will be watching for more of it.
Axel
WHAT GOT ME THINKING
AI on the line at the world's largest tree nursery
SCA has put a vision model to work at Bogrundet, the world's largest forest nursery, where close to 100 million Scots pine, spruce and lodgepole pine seedlings are grown each year. The model looks at each planting cassette and decides in about a second whether it is damaged and should be pulled from the line. It is a quality check that used to rest entirely on the human eye keeping pace.
Axel's notes: When people imagine AI in the forest, they picture a drone taking down a tree. The first place it actually earns its keep is far quieter than that, standing over a conveyor deciding whether a planting cassette is sound enough to keep.
That is exactly the kind of task worth handing to a model. It is repetitive, it is visual, and each call is small but has to be right tens of millions of times a season. A person doing it gets tired and the line slows; a model does not. The value was never going to show up in the dramatic demo. It shows up in the dull job done at scale.
I would not call one cassette check a transformation. But that is the point. The real version of AI in forestry is already running on a conveyor in Norrland, deciding whether a cassette is fit to keep, one small call made correctly all day long. The dull, repetitive jobs are where it lands first, well before the headline robots.
AirForestry fells its first tree by drone in a live forest
AirForestry flew into a real stand and took down a standing tree with a drone, the first time that has happened outside a test site. On its own, the system positioned itself, identified the tree, delimbed it, cut it, and set it down. Back on the test field it has now run the full sequence end to end without a hand on the controls. We covered the company's earlier harvest milestone in March; this is the step from proving the motion to doing it where the trees actually grow.
SkogsAI signs its first customer and launches AnmälansAI
SkogsAI has its first paying customer in Persson Invest Skog. Feedback from Persson Invest Skog AB was used to create AnmälansAI, a sibling of SkogsAI aimed squarely at the felling notification every Swedish harvest has to file.
It pairs a map with an agent that is up to date on the authorities' current guidance, so filing stops being a research project. In beta it was already saving users hours a month.
SLU supercomputer maps 117 million trees in 16 hours
Researchers at SLU segmented every individual tree across 125,000 hectares of northern Sweden, 117 million in total, from dense airborne laser data. They ran 500 parallel processes on the Computational Forestry Lab cluster and finished in 16 hours, work that would take a single desktop about a year. Accuracy for birch, pine and spruce already sits near 90 percent.
Ponsse turns from harvesting to planting with the Buffalo Planter
After four decades building harvesters, the Finnish manufacturer has mounted a planting unit on its Buffalo forwarder frame. One operator can set up to 1,300 seedlings an hour, or about 750 with spot mounding. Ponsse built it with Epec and Novelquip Forestry and is targeting South America first, where large plantations replant millions of identical seedlings on flat ground, the kind of repetitive work a machine handles better than a crew.