Hi Alex,
Do you any thoughts on how much one would have to spend on drones per year to decrease malaria deaths by 1 %? My intuition is that using drones is way less cost-effective than gene-drives.
I'm currently building a $200 drone that kills mosquitoes like a bat. It locates and identifies them with an ultrasonic sonar, and then kills them with a front propeller.
It turns out, the drone part is quite simple. 20cm-wide agile quadcopters can be found in every toy shop. The control theory needed to intercept a target is the same as on an air-to-air missile, so is well-known and documented.
What's hard is building a 3D sonar able to detect and track a mosquito with a high framerate (mosquitoes are quite agile, and when designing interceptors the main criteria for success is your sensor's refresh speed). This is what I'm working on right now!
My sonar uses mass-produced ultrasonic emitters from car park-assist sensors and mass-produced MEMS microphones as ultrasonic receivers. The operating principle is the same as on military radars. It first emits a microsecond-long 40khz ultrasonic pulse. Then, it listens with a 10x10 microphone matrix for echoes and locates them using frequency-domain beamforming. You can find a good explanation on how the receive-side works here : https://www.ribbonfarm.com/2016/06/29/the-daredevil-camera/
I've already built two proof-of-concepts :
The first one I built is able to see the world with a good definition but at a too slow framerate. My writeup made it to the front page of HackerNews and was featured in a hackaday.com article:
The second-one I built uses receive-side beamforming to get a good refresh rate. It is able to scan its environment at 30FPS, which is enough to intercept a mosquito. It can also see sub-centimeter wires. The resolution is not as good as on the first one because I wanted to reduce the design's complexity. Here are two videos of it (color encodes distance to the sensor).
The 3D point cloud might disturb you because ultrasonic waves reflect differently than light, which makes the digital signal processing simpler compared to video at the price of a reduced resolution (which is why bats are more agile than birds that use passive vision)
This sensor will be the basis for the mosquito-hunter drone, but could also be used for robotics in general. It is a a good alternative to lidar with different characteristics (wider field of view, higher framerate and the ability to see very small objects but at a lower resolution).
I'm currently working on a third and final proof-of-concept that combines the resolution of the first sensor with the framerate of the second. Range should be around 40 meters for a 140x140deg field of view.
I've been building this 3d ultrasonic sonar for a few years (cf the article on my blog) and I have been working on my own drones since I’m 14 (visible in the videos I linked above). I also studied the required control theory at school.
https://twitter.com/alextoussss
https://alextoussaint.com/resumev2.pdf
20k€ would let me work on this full-time at least until the end of my gap-year and anything above that would help me buy engineering time to make it faster
I would say a 90% chance of having a viable alternative to LiDAR (really close with my proof-of-concepts, main difficulty is industrialization) and a 50% chance of having a reliable mosquito-hunting machine in the end (still some digital signal processing to figure out, but I'm getting there).
Vasco Grilo
3 days ago
Hi Alex,
Do you any thoughts on how much one would have to spend on drones per year to decrease malaria deaths by 1 %? My intuition is that using drones is way less cost-effective than gene-drives.
Vasco Grilo
3 days ago
@vascoamaralgrilo From here (https://docs.google.com/spreadsheets/d/18ROI6dRdKsNfXg5gIyBa1_7eYOjowfbw5n65zkrLnvc/edit#gid=1364064522), bednets provide 0.546 effective person-years of coverage per $ (= 54.6*10^3/(100*10^3)). So a drone costing 200 $ would have to provide 109 person-years of effective coverage (= 200*0.546) to be competitive. In other words, even ignoring maintenance costs, it would have to protect a family of 5 as effectively as bednets for 21.8 years (= 109/5). This seems super optimistic, so I am confused why people are donating so much to your project.
Alex Toussaint
3 days ago
We’re not targeting the individual person market. We’re targeting communities that could cover whole cities with a few drones. Drones travel fast, so can clear of mosquitoes impressive volumes of air in a short amount of time. Our plan entire densely-populated cities mosquito-free. To give you a rough sense of the figures, a drone covering a 2x2 meters square traveling at 10m/s covers 40 cubic meters of air per second. Over a conservative 10 meters of height, that’s 4 square meters/second. For 1000 drones ($200k capital, less than the typical drone light show) you clear a square kilometer in 250 seconds. With a conservative availability factor of 1/4, you cover a square kilometer or 20k people at Paris density for a million dollars. Assuming a very conservative yearly replacement rate, that’s $50/year/person.
The figure is already good enough for the developed world, and with some economies of scale and more realistic estimations it will well be worth it for the developing world!
Please don’t hesitate if you have any other questions, I’d be happy to answer!
Vasco Grilo
2 days ago
@alextouss Thanks for following up, Alex! Your figure of 50 $/person/year is around half of what I estimated, but it still implies that bednets are much better. People with bednets in DRC (the data I provided above was also for DRC) have a 56 % lower malaria mortality (https://docs.google.com/spreadsheets/d/18ROI6dRdKsNfXg5gIyBa1_7eYOjowfbw5n65zkrLnvc/edit#gid=1364064522&range=B93), so the effective cost for bednets would be 3.27 $/person/year (= 1/0.546/0.56). This means bednets would be 15.3 times (= 50/3.27) as cost-effective as drones even assuming these decrease malaria deaths by 100 % and have no development nor maintenance costs. You can argue your cost would decrease due to economies of scale, but there are also other factors which push it up. For example, I think malaria is more prevalent in rural areas, and rural areas in low income countries are much less dense than Paris.
I do not feel like your numbers are conservative as you say (feel free to link to sources supporting that). You say a drone can cover a square of side of side 2 m, but I assume the rotors do not cover an area of 4 m^2! So you cannot use your linear speed to 10 m/s to do the calculation. Also, 10 m/s (36 km/h) is pretty fast! Would people be happy with having thousands of drones flying at these speed in cities? Would it be safe? I assume drones would need to fly relatively close to people, because that is where the mosquitoes are? Maybe it would be better to just fly the drones in the sources of the mosquitoes (bodies of water, I think), but then gene drives just seem way more effective.
Alex Toussaint
2 days ago
@vascoamaralgrilo Bednets aren’t better, they’re in a different category. While they’re cheap, they don’t cover anything outside of bedtime. Although it’s important everyone in the poorest countries can have one, it won’t be enough to get the last 44% coverage needed, which is what we’re targeting.
About the specific figures, the 2x2m comes from the view distance of the sonar. We’re targeting the sub-250g class of drones, which have very light to no regulation around the world. Your typical football is 320 grams and goes up to 40m/s, so our drones would pack 20x less energy. I’m not concerned about city usage.
Vasco Grilo
2 days ago
@alextouss To clarify, would you agree we should fund whatever is most cost-effective at the margin? If bednets save a life for 5 k$, and drones save a life for 76.5 k$ (= 5*10^3*15.3), then one should fund bednets until their marginal cost to save a life reaches 76.5 k$? GiveWell's current cost-effectiveness bar is 10 times that of unconditional cash transfers (https://www.givewell.org/how-we-work/criteria), which means GiveDirectly still looks 1.53 times (= 15.3/10) as cost-effective as your intervention. GiveDirectly's unconditional cash transfers are supe scalable, and I am pretty confident GiveWell has hundreds of millions of $ of room to fund interventions better than GiveDirectly. So one would need to direct hundreds of millions of $ to GiveWell before your project becomes competitive.
Vasco Grilo
2 days ago
@acx-grants Could you explain your reasoning for donating 20 k$ to this project?
Alex Toussaint
2 days ago
@vascoamaralgrilo I'm not sure I get the reason for your hostility. Your donation criteria is yours, and other people have different ones. If you can't see the value of entirely getting rid of mosquitoes in cities for $50/year/person, than I can't do anything for you.
Austin Chen
about 18 hours ago
I agree with @alextouss; @vascoamaralgrilo, while I appreciate critiques that are intended to help out projects, I think your stance is unnecessarily combative here, or at least shortsighted/missing the point of Manifund. Manifund is primarily a platform for early stage projects, where there is a lot of information value to be gained from people like Alex trying new and novel things and then telling the world about their results. We don't generally expect interventions at these stages to compare favorably to established interventions like bednets on a spreadsheet (though it's a nice benefit if they do!)
I do also think that mosquito gene drives are extremely promising and I would be excited to see project proposals of that sort on Manifund -- if you are yourself interested in working on this or know of people are, encourage them to apply!
Vasco Grilo
about 17 hours ago
@alextouss Sorry for being too negative. I was assuming the goal of this project was preventing malaria deaths more cost-effectively than the most competitive alternatives (like bednets), but I appreciate others may want to fund the project for other reasons (e.g. because it sounds cool).
Thanks for the the feedback, @Austin . If I was proposing a project, I would find it super helpful if someone pointed out to me there are way more cost-effective interventions. I want to contribute to a better world as much as possible, so I would want to know if the money which would support my project could go towards other interventions that would e.g. save more lives. If I was persuaded by the arguments, I would consider halting the project, returning the money to the donors, and then encourage them to donate to a more promising project.
I agree one should account for the value of information of projects. However, do you think anti-mosquito drones have a realistic chance of being more cost-effective than GiveWell's top charities? If not, there is not much relevant information to be gained? Or are you referring to the value of information which is not related to the cost-effectiveness of the intervention, like Alex getting a better picture of what he wants to do in the future? Are there more cost-effective ways of gaining such other information (for example, for e.g. career plans, checking e.g. 80,000 Hours' career guide (https://80000hours.org/career-guide/))?
I understand Manifund has the goal of supporting early stage projects, I agree it makes sense to support novel interventions, and I personally do not think the platform should be restricted to cost-effective projects. However, I think users should also be welcome to point out whether they think a given project is cost-effective or not.
As a side note, I also think GiveWell's top charities are a pretty low bar. I believe the best animal welfare interventions are 1 k times as cost-effective (https://forum.effectivealtruism.org/posts/vBcT7i7AkNJ6u9BcQ/prioritising-animal-welfare-over-global-health-and#Corporate_campaigns_for_chicken_welfare_increase_nearterm_wellbeing_way_more_cost_effectively_than_GiveWell_s_top_charities).
Austin Chen
about 17 hours ago
@vascoamaralgrilo idk, I feel like we're drawing from pretty different worldviews here.
I don't expect people like Alex to look at a complicated calculation that concludes "and therefore gene drives are 10x as effective as anti-mosquito drones" and think "okay, so I should drop what I'm doing and work on gene drives"
Alex might just not believe the calculations -- for plausible reasons! He has a bunch more insight into what's good about drones than an outside observer. I do think anti-mosquito drones have a realistic chance (eg >1%) of being sufficiently cost effective to be part of the fight against malaria
His current set of skills sets him up really well for doing this particular project; founder-market fit is super important in making projects go well
I think you're underrating "because it sounds cool", I guess. To shore this up a bit more, sounding cool is an important factor in getting buzz, being able to raise further funding, get stakeholders to agree to participate, etc.
I think the general approach of analyzing cost effectiveness just doesn't really make sense for projects at this stage, or is at best one of many different lens. Cost effectieness is a reasonable sanity check, maybe; other than that, I'd look to other signals like founder quality and project velocity
Dark shark
8 days ago
Amazing project! This will help so many people!! Sending you love and support from California!
BARROIS
19 days ago
Hello, I am an asitic hornet hunter and iA creator to locate them, do you think that your sound camera can be filtered at their flight frequency (between 100 and 150 Hz) and have a range of 1m.
Guenael Strutt
about 1 month ago
Impressive work, especially on such a budget and tight timeframe!
Can you comment on the presence of turbulence around the drone (which creates noise in the ultrasound spectrum) and the reflectivity of a mosquito (relative to the artifact that you used in the proof of concept). Your approach is sound for 3D scanning in general but I don't see how the SNR numbers add up for the purpose of chasing mosquitoes with a drone.
I have experience in 3D object tracking using ultrasound if you'd like to chat (LinkedIn or email).
Alex Toussaint
about 1 month ago
We've got a 40db SNR from emitter to receiver at 1m with an integration period of 25microseconds.
We have 16 transmitters in a phased array, so with focusing gains we have SNR*16*16.
100 receivers, SNR gains in SNR*100 as noise power adds-up in N and signal power adds-up in N^2
We correlate on a 20 periods-long (17cm) window. Signal adds-up in N^2 and noise in N again.
We're really close to measuring a mosquito's sonar cross section. But, assuming wings 5mm long and 2mm wide, we can estimate it at 10e-6 m^2. A conservative assumption could be around 1e-6 m^2
From the radar range equation : ((10**(40/10))*16*16*100*20*1e-6/((4*3.14)**2))**(1/4) = 2.4 meters.
It turns out, the radar range equation being in R^(1/4) really helps. It stops growing so quickly that even big cross-section differences have a limited effect on the range. Plus, the physics of SNR are really nice with very large transmit and receiver array.
About the drone turbulence I'm not sure it would have a very significant effect in the ultrasonic band we use (40khz). It probably is more of a problem at 20khz than at 40, and there are a lot of ways we could filter it : either electronically, by looking at the spatial and time distribution of the noise, or mechanically by moving it to a region where airflow isn't that disturbed.
The main problem could be the propellers vibrations. But, because we know what the frequencies are (propeller's RPM are controlled by our flight controller) we can dynamically adapt a very narrow cut-band notch filter to ensure it doesn't make its way to our CFAR detection. Flight controllers use the same method to filter propeller noise into gyroscope reading.
Happy to answer any other questions! Invited you on LinkedIn.
Jonas Hyllengren
about 2 months ago
Great work and impressive technology! I'm looking forward to seeing your progress! :)
Alex Toussaint
about 2 months ago
@Jonas-Hyllengren Thanks a lot ! Your contribution will definitely help.
Saul Munn
3 months ago
two main concerns:
is the mosquito's death painful?
would this have harmful consequences down the line? (e.g. the frogs starve because they don't have enough food)
otherwise, this seems like a really neat idea!
Alex Toussaint
about 2 months ago
Hard to be sure about it :/
That’s already been studied, and the general consensus seems to be that mosquitoes are a small part of the alimentation of their predators, while they cause a disproportionate amount of harm to the human population.
David Kasten
3 months ago
I have no clue whether or not this will work, but as one of those unfortunates who is disproportionately attractive to mosquitoes and gets bit a lot each summer, I want to praise anyone that is attempting to MURDER THOSE BUGS.
If this worked (and was compliant with DC restrictions on drone use), I would immediately purchase one for my backyard at a $1000 price point. Wouldn't even blink.
Alex Toussaint
3 months ago
@davekasten We’re mainly targeting the developing world for the moment, but there’s no reason for us to restrict our reach! Thank you for your support.