Comment by alan-crowe | Hacker News Reader

Comment by alan-crowe | Hacker News Reader

Parent Full thread

My reference point for this kind of thing is the video cassette recorder (VCR). Trusting to memory I think that they entered the consumer market at around $700 and were down to around $100 by the time they were obsolete. But before they reached the consumer market, they were a super-expensive professional item. Getting helical scan VCRs (a high precision, complicated electro-mechanical technology) to the consumer market involved impossible price reductions, just to reach the consume price point of $700. I never expected further big price reductions to follow and was wrong.

A weed killing laser robot seems obviously stupid - an expensive electro-mechanical system generating trivial value with each weed zapped. There is no way to get the price down low enough for this to make sense. But CD players have servo control of the tracking and focus of the laser. Much lower power than needed for weed wacking, but much faster and more accurate than needed and very cheap. https://www.youtube.com/watch?v=wplLkxBsqQo Adequate cameras are cheap enough. AI vision processing is on a roll and we can look forward to huge price falls. The technology looks promising from the outside.

My reference point is electric cars. They were a complete non-starter, all but physically impossible because the energy density of batteries was so incredibly much lower than that of gas.

Then batteries got better, and hybrids came out that had dramatically better mileage than ICE cars. But they were expensive, and looked weird, and electric cars still weren't possible.

Then electric cars came out that were expensive, and low range, but pretty viable for city driving. But the range would never match ICE cars, and they'd always be to expensive.

We're now at a point where the infrastructure and economics hasn't caught up, and the range is still arguably a little low for super long road trips, but once we get working fast chargers saturating most transit corridors, and once we get cheaper, smaller cars offered, we'll be pretty much there.

From "The physics really just don't seem to add up" to "Well they're a little pricey" in, what, thirty years?

I honestly can’t imagine a future where we don’t have robots tending to each individual plant coming out of the ground like that one plant is the most important thing in the world.

We currently treat an entire field as one uniform thing because that’s the cheapest way to do it. We plant in rows to allow our simple machines to operate on them. We drive massive combines up and down the field. We spray the entire field uniformly.

I think there’s a move in the precision farming world to using remote sensing imaging to treat smaller tracts of land (like an acre) separately. But at some point the technology will be cheap enough to focus in on each individual plant.

Clearly there’s some extra growth that each plant would have if it would treated really well and harvested at the perfect time. That’s the opportunity. The cost is developing swarms of robots that go up and down the fields 24 hours a day checking on soil moisture, leaf disease, weeds and doing what it needs to tend to that plant. I think the hardware already exists and is cheap enough. It’s now just a matter of developing the software (yes, the devil is in the word “just”).

If there are 10 plants per square meter and we give the robot 1 minute per plant, and there are 10,000 square meters in a hectare, then we will need 70 robots to tend an entire hectare in a single 24 hour period. That’s the sort of math that I’ve been thinking of.

For well over 10 years it has been very common to vary treatment within a field. Any harvester made in that time has had real-time GPS tracking of the yield throughout the field. Spray applications can be applied to particular areas with an indicated need, with individual sprayer head control (all GPS tracked). Year-over-year tracking enables another level of improvement. Pairing satellite or drone imagery is also common to determine what the best move is going forward.

A decade ago I made a spreadsheet to simulate the robot tending to a field problem, working with an entrepreneur. In the end, I couldn't find a wide range of parameters where it could make sense for corn or soybeans. The robots would simply be too tiny, slow, and expensive to be feasible.

Some pros the entrepreneur was targeting were reduced soil compaction (leading to better yield), ability to operate in more marginal conditions (getting stuck in mud or slipping/spinning wheels messes up fields), and ability to deploy a fleet of robots rather than being constrained to working one part of a field in a huge piece of equipment.

Reply on news.ycombinator.com