Physicists know that nature is lazy. The crops which will give the most beneficial return for the farmer are probably not the plants best adapted to survive the environment, and are unlikely to consistently win the never-ending competition over space, nutrients, water and sunlight with the other plants around it. The weeds will win.
But naturalists know that all plants want to survive, and all that is needed are mechanisms to give sufficient advantage to to desired plants to enable to take the advantage over the weeds (and hopefully over their parasites too). Farming is not about completely eradicating all weeds and parasites, but simply to disrupt them enough for the crops to get the advantage over the weeds, and using their natural defences against parasites.
Weed control mechanisms fit into two categories - general and specific.
General: apply a process to the whole of the crop, and hope that the mechanism applied specifically harms weeds more than the crop.
Specific: apply a mechanism specifically to plants identified as weeds, and apply a general disruption mechanism to harm those specific weeds more that the crop plants around it.
(The labels are confusing and often used in the reverse sense - a specific herbicide is applied generally to the crop in the hope that it will damage the weeds more than the crop).
A special note about specific and general herbicides is appropriate here.
Normally farmers use massive booms for general spraying of pesticides (and fertilisers). Humans are expensive and large tractors, while extremely expensive, are actually cost effective if labour costs are high. However, automation can work on a much smaller scale.
This is a small spraying rig designed to be towed behind a small garden tractor. It can spray pesticide (or liquid fertiliser such as Seasol) over a large area, albeit slowly.
Spreading liquid evenly over a large area is actually quite a tricky thing to do - it involves careful flow-rate calculations and, to do the job properly, probably several passes over the same area to even the distribution, and sometimes even carefully marking of areas already covered.
Incorrect application amounts are about more than just wastage. Some pesticides are very harmful to crops or livestock in larger quantities, and also risk polluting the environment.
A spray trailer could be attached the rover, and it could be used to spray the house paddock, or the other paddocks. This would have several possible benefits.
A human doesn't need to breathe in the chemical being sprayed (though there is still the issue with residual chemical being left in soil, plants, food, and the environment).
The chemical could be applied more consistently - as the rover path and rate of spray can be adjusted by more accurate means that human estimation and memory.
This is a small spreader rig designed to be towed behind a small garden tractor. It spreads fertiliser or coarse seeds over a large area.
Solids too are hard to spread evenly.
Normally both of these would be considered to too small for a farm because of the excessive time required sitting on a tractor - the most expensive thing to buy is human labour. However they are ideal applications for an autonomous rover. The rover can simply tow the applicators over a pre-determined course - as fast, as slow, and as often as required.
In the case of the spreader, the rate of use is determined by the rate of wheel turn (and the manual rate lever, which could be replaced by an actuator In the case of the spray rig, the flow rate can be adjusted for speed (as determined by a GPS) by control of the electric pump.
While agricultural robotics can be useful in either general or specific weeding applications, the most promise is in specific control mechanisms. An autonomous tractor roving the field spraying chemical over all the crop may be a way to produce a crop more cost effectively (saving the wages of the tractor operator), but it is unlikely to result in less chemical (or less environmentally damaging chemicals) being used. Specific control mechanisms on the other hand promise reduced chemical use.
Employing a small army of farm workers to remove weeds with a hoes is prohibitively expensive for most of the developed world's farms. Employing a small army of robots to remove weeds with a hoe like disruptor is quite plausible
Even employing a small army of farm workers to specifically spray weeds with a the smallest amount of chemical to kill that weed is prohibitively expensive for most of the world's farms. Employing a small army of robots to spray small amounts of herbicide into plants identified as weeds is quite plausible
Specific control is about stressing a weed, and making sure it is more stressed than the crop plants around it. Stressing might be done by
However, for any specific mechanism to be effective it must be able to specifically target the weeds. This can be done either by dead reckoning or identification of weeds.
Grid-based dead reckoning is the mechanism employed by several system including Goat Industries' Weedinator. If you know the location of every plant in your crop (even if it is just the knowledge of them being in a grid with known separation and starting point), then disrupting everything in between the rows and columns (with a big egg-beater) will kill the weeds. There are also larger tractor-pulled machines which do similar things on a larger scale.
Individual identification can take at least two approaches.
Some people might object to the notion of getting a robot to spray herbicide at all, based on an argument "we should work toward a solution which does away with potentially harmful chemicals altogether". This is admirable in intent, but not necessarily a realistic short or even medium-term goal.
Spot-spraying a general herbicide (such as Glyphosate) simply means less chemical sprayed around the place than blanketing a field with a specific herbicide. Glyphosate use is controversial, but the rational arguments against it are actually against its excessive use (not because of the excessive harmfulness of the chemical itself). Granted, there is no such thing as a good herbicide, but some are worse than others, and using less of any herbicide is a good thing - which is precisely what we are trying to achieve here.
A Google search reveals several interesting alternatives to spraying herbicide.
Blasting the weeds briefly with microwaves to kill it (this seems a little high energy for a battery-powered rover).
Hitting the weed with a punch, and effectively stamping it into the ground (sorry, couldn't find a picture).
Disrupting the weed's centre with an digging auger.
This seems straightforward, but actually involves building a sturdy three-degrees-of-freedom mechanism (four counting turning the auger).
Develop a 'squirt' to aim a nozzle at a weed and spray it with a small (very carefully controlled) burst of herbicide (anyone with any experience with this is invited to write to me).