science
Auto Added by WPeMatico
StarCraft II-playing AI AlphaStar takes out pros undefeated
Losing to the computer in StarCraft has been a tradition of mine since the first game came out in 1998. Of course, the built-in “AI” is trivial for serious players to beat, and for years researchers have attempted to replicate human strategy and skill in the latest version of the game. They’ve just made a huge leap with AlphaStar, which recently beat two leading pros 5-0.
The new system was created by DeepMind, and in many ways it’s very unlike what you might call a “traditional” StarCraft AI. The computer opponents you can select in the game are really pretty dumb — they have basic built-in strategies, know in general how to attack and defend, and how to progress down the tech tree. But they lack everything that makes a human player strong: adaptability, improvisation, and imagination.
AlphaStar is different. It learned from watching humans play at first, but soon honed its skills by playing against facets of itself.
The first iterations watched replays of games to learn the basics of “micro” (i.e. controlling units effectively) and “macro” (i.e. game economy and long-term goals) strategy. With this knowledge it was able to beat the in-game computer opponents on their hardest setting 95 percent of the time. But as any pro will tell you, that’s child’s play. So the real work started here.
Hundreds of agents were spawned and pitted against each other.
Because StarCraft is such a complex game, it would be silly to think that there’s an single optimal strategy that works in all situations. So once the machine learning agent was essentially split into hundreds of versions of itself, each given a slightly different task or strategy. One might attempt to achieve air superiority at all costs; another to focus on teching up; another to try various “cheese” attempts like worker rushes and the like. Some were even given strong agents as targets, caring about nothing else but beating an already successful strategy.
This family of agents fought and fought for hundreds of years of in-game time (undertaken in parallel, of course). Over time the various agents learned (and of course reported back) various stratagems, from simple things such as how to scatter units under an area-of-effect attack to complex multi-pronged offenses. Putting them all together produced the highly robust AlphaStar agent, with some 200 years of gameplay under its belt.
Most StarCraft II pros are well under 200, so that’s a bit of an unfair advantage. There’s also the fact that AlphaStar, in its original incarnation anyway, has two other major benefits.
First, it gets its information directly from the game engine, rather than having to observe the game screen — so it knows instantly that a unit is down to 20 HP without having to click on it. Second, it can (though it doesn’t always) perform far more “actions per minute” than a human, because it isn’t limited by fleshy hands and banks of buttons. APM is just one measure among many that determines the outcome of a match, but it can’t hurt to be able to command a guy twenty times in a second rather than two or three.
It’s worth noting here that AIs for micro control have existed for years, having demonstrated their prowess in the original StarCraft. It’s incredibly useful to be able to perfectly cycle out units in a firefight so none takes lethal damage, or to perfectly time movements so no attacker is idle, but the truth is good strategy beats good tactics pretty much every time. A good player can counter the perfect micro of an AI and take that valuable tool out of play.
AlphaStar was matched up against two pro players, MaNa and TLO of the highly competitive Team Liquid. It beat them both handily, and the pros seemed excited rather than depressed by the machine learning system’s skill. Here’s game 2 against MaNa:
In comments after the game series, MaNa said:
I was impressed to see AlphaStar pull off advanced moves and different strategies across almost every game, using a very human style of gameplay I wouldn’t have expected. I’ve realised how much my gameplay relies on forcing mistakes and being able to exploit human reactions, so this has put the game in a whole new light for me. We’re all excited to see what comes next.
And TLO, who actually is a Zerg main but gamely played Protoss for the experiment:
I was surprised by how strong the agent was. AlphaStar takes well-known strategies and turns them on their head. The agent demonstrated strategies I hadn’t thought of before, which means there may still be new ways of playing the game that we haven’t fully explored yet.
You can get the replays of the matches here.
AlphaStar is inarguably a strong player, but there are some important caveats here. First, when they handicapped the agent by making it play like a human, in that it had to move the camera around, could only click on visible units, had a human-like delay on perception, and so on, it was far less strong and in fact was beaten by MaNa. But that version, which perhaps may become the benchmark rather than its untethered cousin, is still under development, so for that and other reasons it was never going to be as strong.
AlphaStar only plays Protoss, and the most successful versions of itself used very micro-heavy units.
Most importantly, though, AlphaStar is still an extreme specialist. It only plays Protoss versus Protoss — probably has no idea what a zerg looks like — with a single opponent, on a single map. As anyone who has played the game can tell you, the map and the races produce all kinds of variations which massively complicate gameplay and strategy. In essence, AlphaStar is playing only a tiny fraction of the game — though admittedly many players also specialize like this.
That said, the groundwork of designing a self-training agent is the hard part — the actual training is a matter of time and computing power. If it’s 1v1v1 on Bloodbath maybe it’s stalker/zealot time, while if it’s 2v2 on a big map with lots of elevation, out come the air units. (Is it obvious I’m not up on my SC2 strats?)
The project continues and AlphaStar will grow stronger, naturally, but the team at DeepMind thinks that some of the basics of the system, for instance how it efficiently visualizes the rest of the game as a result of every move it makes, could be applied in many other areas where AIs must repeatedly make decisions that affect a complex and long-term series of outcomes.
Powered by WPeMatico
Sustainable Ocean Alliance nets $1.5 million donation from Benioffs
Healthy oceans are on the minds of Marc and Lynne Benioff, and they showed it today with a $1.5 million donation to the Sustainable Ocean Alliance (SOA), a new nonprofit attempting to promote and incubate conservation-focused startups. The money will considerably expand the organization’s upcoming Ocean Solutions accelerator.
Benioff appeared Wednesday on a panel at Davos about the “ocean economy,” at which he mentioned the donation and SOA. He joined rather a powerhouse lineup to address the issues of environmental dangers threatening wallets as well as whales: Michelle Bachelet (U.N. High Commissioner for Human Rights), Enric Sala (an Explorer-in-Residence at National Geographic), Nina Jensen of REV Ocean and indefatigable environmental crusader Al Gore. I certainly wish I could have attended.
It’s clear that the Salesforce founder is as concerned about environmental issues as he is about social ones, and as ready to write a check when there’s a compelling reason to do so.
Benioff at Disrupt SF in 2016
“Our oceans are in grave danger, due to the many consequences of climate change and pollution,” he said in a press release announcing the donation. “These challenges can be solved with investment and innovation. Lynne and I are proud to support [SOA founder] Daniela Fernandez and the Sustainable Ocean Alliance’s bold vision to create 100 new startups by 2021 to help heal the ocean.”
The SOA started its accelerator last year with a handful of interesting ocean- and conservation-focused startups: a device to keep fish from getting tangled in nets, wave-harvesting energy tech, materials for oil cleanups, that sort of thing. It’s got another batch planned and the Benioff’s donation will allow it to triple the number of startups included. Several will be going to the “Accelerator at Sea,” an eight-day event aboard a Lindblad Expeditions ship sailing from Alaska this summer.
Last year the organization also got a sudden cash infusion from a motivated donor: the mysterious Pine, who distributed some $86 million to charity (and nonprofits like SOA) after making a tremendous amount of money on Bitcoin. These are one-off donations, naturally — so of course financial sustainability, as well as ecological, is on Fernandez’s mind.
“We realize that we cannot simply depend on individual donors or anonymous cryptocurrency gifts. We have had difficulty finding traditional forms of funding for SOA due to the limited amount of funds that are allocated to such a niche sector,” Fernandez, who is at Davos but unfortunately not on the aforementioned panel, told me.
“Instead of only having to fundraise, we have had to create new funders by educating them about the importance of protecting the ocean. It is the typical entrepreneurial scenario of building the plane while flying it. However, in our case, we had to build the plane while simultaneously developing the aircraft market.”
As part of that the nonprofit now plans to release a yearly “State of Our Ocean” report — the first came out today. It’s not so much a scholarly or analytical report like you might have from NOAA or national fisheries or wildlife concerns. Fernandez says this one “takes into account the perspective of young people who are on the ground working to solve the issues at hand. SOA interviewed 3,000 young ocean leaders from around the world who gave their input as to what the ocean priorities should be in 2019 and graded our current world leaders on their efforts to restore the health of the ocean.”
It’s good to ask the un-jaded youngs about things like this, and SOA specifically aims to find and promote young entrepreneurs and activists, so it’s on brand. I’ve read through it and there’s a lot of info about impending disasters, many of which have to do with climate change, but plenty are caused by people as well (or rather, caused by people more recently). It’s a bit depressing, but what isn’t?
Hopefully the cash infusion will help scoop up more of those motivated young folks into the program. We’ll probably hear more from the SOA when it finds some more startups to load into the accelerator.
Powered by WPeMatico
Driving down the cost of preserving genetic material, Acorn Biolabs raises $3.3 million
Acorn Biolabs wants consumers to pay them to store genetic material in a bet that the increasing advances in targeted genetic therapies will yield better healthcare results down the line.
The company’s pitch is to “Save young cells today, live a longer, better, tomorrow.” It’s a gamble on the frontiers of healthcare technology that has managed to net the company $3.3 million in seed financing from some of Canada’s busiest investors.
For the Toronto-based company, the pitch isn’t just around banking genetic material — a practice that’s been around for years — it’s about making that process cheaper and easier.
Acorn has come up with a way to collect and preserve the genetic material contained in hair follicles, giving its customers a way to collect full-genome information at home rather than having to come in to a facility and getting bone marrow drawn (the practice at one of its competitors, Forever Labs) .
“We have developed a proprietary media that cells are submerged in that maintains the viability of those cells as they’re being transported to our labs for processing,” says Acorn Biolabs chief executive Dr. Drew Taylor.
“Rapid advancements in the therapeutic use of cells, including the ability to grow human tissue sections, cartilage, artificial skin and stem cells, are already being delivered. Entire heart, liver and kidneys are really just around the corner. The urgency around collecting, preserving and banking youthful cells for future use is real and freezing the clock on your cells will ensure you can leverage them later when you need them,” Taylor said in a statement.
Typically, the cost of banking a full genome test is roughly $2,000 to $3,000, and Acorn says they can drop that cost to less than $1,000. Beyond the cost of taking the sample and storing it, Acorn says it will reduce to roughly $100 a year the fees to store such genetic materials.
It’s important to note that healthcare doesn’t cover any of this. It’s a voluntary service for those neurotic enough or concerned enough about the future of healthcare and their potential health.
There’s also no services that Acorn will provide on the back end of the storage… yet.
“What people do need to realize is that there is power with that data that can improve healthcare. Down the road we will be able to use that data to help people collect that data and power studies,” says Taylor.
The $3.3 million the company raised came from Real Ventures, Globalive Technology, Pool Global Partners and Epic Capital Management and other undisclosed investors.
“Until now, any live cell collection solutions have been highly expensive, invasive and often painful, as well as being geographically limited to specialized clinics,” said Anthony Lacavera, founder and chairman at Globalive. “Acorn is an industry-leading example of how technology can bring real innovation to enable future healthcare solutions that will have meaningful impact on people’s wellbeing and longevity, while at the same time — make it easy, affordable and frictionless for everyone.”
Powered by WPeMatico
Geoengineering could solve our climate problems if anyone allowed it
This weekend, I finished reading Oliver Morton’s The Planet Remade (thanks to reader Eliot Peper for recommending it). Morton has a multitude of goals with the book, but there were two I think are deeply valuable. First, geoengineering is a plausible approach to solving our climate problems this century, and second, engineering the climate generates tough policy challenges, but also opportunities to make the planet more equitable.
TechCrunch is experimenting with new content forms. This is a rough draft of something new — provide your feedback directly to the author (Danny at danny@techcrunch.com) if you like or hate something here.
First and foremost: the book is mind-expanding in the best way possible. Morton confronts an extremely contentious issue with judicious facts and supreme insight gleaned over many years of studying geoengineering. Whether you are a dedicated acolyte of cloud seeding and veils or a committed opponent to any tampering of earth’s environment, he has developed a book that forces us to think about our actions and ultimately what the consequences of those choices are.
Frankly, those choices offer stark consequences. Morton describes the challenge of climate this century:
The world’s population is expected to grow from seven billion today to more or less ten billion by 2100. By that time the number of people enjoying rich-world energy privileges should also reach ten billion. So the challenge is to achieve for an extra eight billion people in the twenty-first century what was achieved for two billion in the twentieth century. Meeting that challenge implies a lot more energy usage.
Morton is a staunch environmentalist and deeply concerned about environmental justice and the inequities of the planet. But he is also a “climate realist” — he understands that our current solutions to climate change are not really solutions at all, since they either lack the scale required to solve the problem, or will continue to exacerbate existing inequities between different people of this planet.
For example, take emissions-free nuclear power, which is brought up as a panacea to our fossil fuel-driven economy. Morton writes:
If the world had the capacity to deliver one of the largest nuclear power plants ever built once a week, week in and week out, it would take 20 years to replace the current stock of coal-fired plants (at present, the world builds about three or four nuclear power plants a year, and retires old ones almost as quickly).
Sure, nuclear power plants are a literal solution, but most definitely not a pragmatic one since the scale required is just not there.
He also spends significant time deconstructing recent climate negotiations, finding that the focus on carbon has been something of a red herring (many other emissions are far worse than carbon and less directly connected to the modern industrial economy). Instead, they have been driven by the alignment of different environmentally-concerned parties:
Carbon dioxide suited scientists because it seemed like a straightforward measure of the problem. It suited greens because it was a pretty good proxy for the industrial society against which their movement was a reaction. The international negotiations that set up the UNFCCC showed that it suited developing countries because it was primarily a developed-country issue; at the time of Rio, the vast majority of all the industrial emissions since the the eighteenth century had come from Europe and America.
Carbon is of course a problem, but it has become a tagline, a brand, a cri de coeur of the international climate movement. Yet the challenges facing the planet are so much deeper than just carbon.
To avoid that narrow focus, Morton argues for a complete reframing of the climate debate toward solutions that can actually repair the climate, and even improve it for diverse populations around the world.
Now, the term “geoengineering” brings with it a bag of Hollywood-induced imagery of nuclear winters and globe-spanning hurricanes. Morton addresses those risks across his chapters, noting that geoengineering can indeed go wrong.
Even so, he convincingly argues that there are geoengineering techniques designed around key climate processes that can be high leverage, reversible, testable, and that have the scale required to actually solve climate challenges in a sustainable way. These processes aren’t speculation — we (mostly) understand the science today, and have pathways toward the technology required to execute a strategy.
The real challenge — as it always is — are humans and their governments. Morton notes that climate change has a huge deleterious impact on nations such as Maldives, but that it can also benefit certain regions by transitioning them from colder to more temperate climates.
That means that any geoengineering solution is going to face the prospect of creating winners and losers. Any international agreement is going to have to contend with those politics, and design mechanisms to ameliorate their effects.
Much as Morton calls for a planet remade, he sees an opportunity for geoengineering to trigger reflection among governments on their own interests:
Much better, rather than treating geoengineering as a technocratic way of avoiding politics, to use it as a way of reinventing politics. Exploring the potential of geoengineering could spur and shape the development of a new way of making planetary decisions. The aim should not be the development of a thermostat alone; it should be the development of a new hand to use it.
Environmentalists may balk at the idea of allowing humans to have their hands on any part of the earth system. But we are here, all seven billion of us, and we already have our brutal hands on the system. The question is whether we can start to use our hands in a far more productive way that can make the earth sustainable for centuries to come. As Morton notes, “The planet has been remade, is being remade, will be remade.” Geoengineering technologies offer solutions, if we can agree in how to use them.
Share your feedback on your startup’s attorney
My colleague Eric Eldon and I are reaching out to startup founders and execs about their experiences with their attorneys. Our goal is to identify the leading lights of the industry and help spark discussions around best practices. If you have an attorney you thought did a fantastic job for your startup, let us know using this short Google Forms survey and also spread the word. We will share the results and more in the coming weeks.
Stray Thoughts (aka, what I am reading)
Short summaries and analysis of important news stories
Why Gutenberg can still recognize the book
Craig Mod wrote a compelling piece in Wired on the future of the book, and why today’s books essentially look the same as when the printing press was first invented. Despite the prognosticators expecting books to have moving pictures, interactivity, and dynamic narratives, almost nothing in that direction has actually occurred as readers continue to enjoy the traditional format. Instead, where the real innovation has taken place is on the business side, where new models from crowdfunding to email subscriptions have transformed the economics of book publishing.
Automattic’s Newspack to drive revenue for smaller publishers
While content management systems have been around for decades, almost none of these systems are designed to create revenues for their users out of the box. WordPress doesn’t have any subscription features or advertising networks built-in, which means that sites that want to make money have to spend a lot of dollars just to get setup and started.
So the announcement this morning that Automattic, the owner of WordPress.com, is going to offer a new platform combining content management with revenue called Newspack is both interesting and definitely needed. It’s a proper extension of their existing platform, and a reminder for product managers that the sustainability of their customers is critical for long-term success.
Huawei sales executive arrested in Poland
We have been following Huawei’s travails in the West for some time. One major point of contention is whether the company spies on behalf of the Chinese government. Western governments have argued that it does, but as China has repeatedly noted, they have never provided any proof.
On Friday in Poland, a Huawei executive was arrested for alleged espionage, which could provide the first public evidence of collusion between Huawei and Beijing. The company subsequently fired the executive and claimed that his actions were unrelated to the company. Poland has since called on NATO countries to remove Huawei equipment from their telecommunications infrastructure. Huawei equipment is widely installed in Europe and European governments have so far evaded calls by the U.S. to boycott the company. As the largest telecom equipment manufacturer in the world, Huawei’s response could have vast repercussions for the deployment of 5G networks.
PG&E – oh boy
Silicon Valley’s (and much of California’s) gas and electric utility is going bankrupt following massive liability claims against the utility due to its equipment sparking wildfires over the past few years. California may lead the world in innovation, but it seems to always be on the precipice of disaster when it comes to infrastructure.
What’s next & obsessions
- I am reading The Color of Law by Richard Rothstein
- Arman and I are interested in societal resilience startups that are targeting areas like water security, housing, infrastructure, climate change, disaster response, etc. Reach out if you have ideas or companies here.
Powered by WPeMatico
This project is mapping every solar panel in the country using machine learning
Renewable energy is the future, but at present no one is tracking just who’s got solar panels on their roof, in their back yard, or a shared neighborhood installation. Fortunately, solar panels generally work best when exposed to the light. That makes them easy to spot, and count, from orbit — which is just what the DeepSolar project is doing.
There are a number of initiatives for collecting this information — some regulated, some voluntary, some automated. But none of them is comprehensive enough or accurate enough to base policy or business decisions on at a national or state level.
Stanford engineers (mechanical and civil, respectively) Arun Majumdar and Ram Rajagopal decided to remedy this with what seems like, in retrospect, rather an obvious solution.
Machine learning systems are great at looking at images and finding objects they’ve been “trained” to recognize, whether it’s cats, faces, or cars… so why not solar panels?
Their team, including grad students Jiafan Yu and Zhecheng Wang, put together an image recognition machine learning agent trained on hundreds of thousands of satellite images. The model learns both to identify the presence of solar panels in an image, and to find the shape and area of those panels.
Having evaluated the model on nearly a hundred thousand other randomly sampled satellite images of the U.S., they found they achieved an accuracy of about 90 percent (slightly more or less depending on how it’s measured), which is well ahead of other models, and it estimated cell size with only about a 3 percent error. (Its main weakness is very small installations, Rajagopal told me, but this is partially due to the limits of the imagery.)
The team then put the model to work chewing through over a billion image tiles covering as much of the lower 48 states as they could find suitable imagery for. That excludes quite a bit of area, but consider that much of that is, for example, mountains. Not a lot of solar installations there, and few people are trying to put up cells in national parks.
All in all it’s about 6 percent of the actual country — but Rajagopal pointed out that urban areas comprise only about 3.5 percent, so this covers all of them and more. He estimated that perhaps perhaps 5 percent of installations are in the areas the system has yet to process (but is working on).
Scanning took a whole month, but at the end the model had found 1.47 million individual solar installations (which could be a few panels on a roof or a whole solar farm). That’s many more than have been counted by other efforts, and the most successful of those didn’t come with the exact location, as DeepSolar’s data does.
Basic plotting of this data produces all kinds of interesting new info. You can compare solar installation density at the state, county, census tract, or even square mile level and compare that to all kinds of other metrics — average sunny days per year, household income, voting preference, and so on.
A couple interesting findings: Only 4 percent of all census tracts (roughly 3,000 out of 75,000) had more than 100 residential-scale solar systems, meaning installations are highly concentrated. Residential solar made up 87 percent of the total installation count, but with a median size of around 25 square meters, only 34 percent of the total solar cell surface area.
Peak deployment density can be found where there are about a thousand people per square mile — think a small town or suburb, not a major city. And there’s a sort of inflection point at which people start installing: when an area receives more than 4.5 kWh per square meter per day of solar radiation. How that corresponds to weather, location, exposure and so on is a more complicated question.
This and other demographics are all good information to know if you want to invest in solar, since they basically tell you where it’s justified or needed.
“We have created and released a website where you can play with the data at the aggregated level (we are keeping it at census tract level) to respect the privacy of consumers,” Rajagopal said. “We are exploring how to make individual detections public while respecting privacy (perhaps by encouraging public participation and crowdsourcing).”
“We decided to share all of the work in open source to encourage others in industry and academia to utilize both the method as well as the data to produce more insights. We feel that changes need to happen fast, and this is one of the ways to aid in that. Perhaps in the future, services can be built around this type of data,” he continued.
Plans are underway to expand the service to the rest of the U.S. and other countries as well. The data is available to peruse here, or here as a map; the team’s paper describing the project was published today in the journal Joule.
Powered by WPeMatico
That night, a forest flew: DroneSeed is planting trees from the air
Wildfires are consuming our forests and grasslands faster than we can replace them. It’s a vicious cycle of destruction and inadequate restoration rooted, so to speak, in decades of neglect of the institutions and technologies needed to keep these environments healthy.
DroneSeed is a Seattle-based startup that aims to combat this growing problem with a modern toolkit that scales: drones, artificial intelligence and biological engineering. And it’s even more complicated than it sounds.
Trees in decline
A bit of background first. The problem of disappearing forests is a complex one, but it boils down to a few major factors: climate change, outdated methods and shrinking budgets (and as you can imagine, all three are related).
Forest fires are a natural occurrence, of course. And they’re necessary, as you’ve likely read, to sort of clear the deck for new growth to take hold. But climate change, monoculture growth, population increases, lack of control burns and other factors have led to these events taking place not just more often, but more extensively and to more permanent effect.
On average, the U.S. is losing 7 million acres a year. That’s not easy to replace to begin with — and as budgets for the likes of national and state forest upkeep have shrunk continually over the last half century, there have been fewer and fewer resources with which to combat this trend.
The most effective and common reforestation technique for a recently burned woodland is human planters carrying sacks of seedlings and manually selecting and placing them across miles of landscapes. This back-breaking work is rarely done by anyone for more than a year or two, so labor is scarce and turnover is intense.
Even if the labor was available on tap, the trees might not be. Seedlings take time to grow in nurseries and a major wildfire might necessitate the purchase and planting of millions of new trees. It’s impossible for nurseries to anticipate this demand, and the risk associated with growing such numbers on speculation is more than many can afford. One missed guess could put the whole operation underwater.
Meanwhile, if nothing gets planted, invasive weeds move in with a vengeance, claiming huge areas that were once old growth forests. Lacking the labor and tree inventory to stem this possibility, forest keepers resort to a stopgap measure: use helicopters to drench the area in herbicides to kill weeds, then saturate it with fast-growing cheatgrass or the like. (The alternative to spraying is, again, the manual approach: machetes.)
At least then, in a year, instead of a weedy wasteland, you have a grassy monoculture — not a forest, but it’ll do until the forest gets here.
One final complication: helicopter spraying is a horrendously dangerous profession. These pilots are flying at sub-100-foot elevations, performing high-speed maneuvers so that their sprays reach the very edge of burn zones but they don’t crash head-on into the trees. This is an extremely dangerous occupation: 80 to 100 crashes occur every year in the U.S. alone.
In short, there are more and worse fires and we have fewer resources — and dated ones at that — with which to restore forests after them.
These are facts anyone in forest ecology and logging are familiar with, but perhaps not as well known among technologists. We do tend to stay in areas with cell coverage. But it turns out that a boost from the cloistered knowledge workers of the tech world — specifically those in the Emerald City — may be exactly what the industry and ecosystem require.
Simple idea, complex solution
So what’s the solution to all this? Automation, right?
Automation, especially via robotics, is proverbially suited for jobs that are “dull, dirty, and dangerous.” Restoring a forest is dirty and dangerous to be sure. But dull isn’t quite right. It turns out that the process requires far more intelligence than anyone was willing, it seems, to apply to the problem — with the exception of those planters. That’s changing.
Earlier this year, DroneSeed was awarded the first multi-craft, over-55-pounds unmanned aerial vehicle license ever issued by the FAA. Its custom UAV platforms, equipped with multispectral camera arrays, high-end lidar, six-gallon tanks of herbicide and proprietary seed dispersal mechanisms have been hired by several major forest management companies, with government entities eyeing the service as well.
These drones scout a burned area, mapping it down to as high as centimeter accuracy, including objects and plant species, fumigate it efficiently and autonomously, identify where trees would grow best, then deploy painstakingly designed seed-nutrient packages to those locations. It’s cheaper than people, less wasteful and dangerous than helicopters and smart enough to scale to national forests currently at risk of permanent damage.
I met with the company’s team at their headquarters near Ballard, where complete and half-finished drones sat on top of their cases and the air was thick with capsaicin (we’ll get to that).
The idea for the company began when founder and CEO Grant Canary burned through a few sustainable startup ideas after his last company was acquired, and was told, in his despondency, that he might have to just go plant trees. Canary took his friend’s suggestion literally.
“I started looking into how it’s done today,” he told me. “It’s incredibly outdated. Even at the most sophisticated companies in the world, planters are superheroes that use bags and a shovel to plant trees. They’re being paid to move material over mountainous terrain and be a simple AI and determine where to plant trees where they will grow — microsites. We are now able to do both these functions with drones. This allows those same workers to address much larger areas faster without the caloric wear and tear.”
It may not surprise you to hear that investors are not especially hot on forest restoration (I joked that it was a “growth industry” but really because of the reasons above it’s in dire straits).
But investors are interested in automation, machine learning, drones and especially government contracts. So the pitch took that form. With the money DroneSeed secured, it has built its modestly sized but highly accomplished team and produced the prototype drones with which is has captured several significant contracts before even announcing that it exists.
“We definitely don’t fit the mold or metrics most startups are judged on. The nice thing about not fitting the mold is people double take and then get curious,” Canary said. “Once they see we can actually execute and have been with 3 of the 5 largest timber companies in the U.S. for years, they get excited and really start advocating hard for us.”
The company went through Techstars, and Social Capital helped them get on their feet, with Spero Ventures joining up after the company got some groundwork done.
If things go as DroneSeed hopes, these drones could be deployed all over the world by trained teams, allowing spraying and planting efforts in nurseries and natural forests to take place exponentially faster and more efficiently than they are today. It’s genuine change-the-world-from-your-garage stuff, which is why this article is so long.
Hunter (weed) killers
The job at hand isn’t simple or even straightforward. Every landscape differs from every other, not just in the shape and size of the area to be treated but the ecology, native species, soil type and acidity, type of fire or logging that cleared it and so on. So the first and most important task is to gather information.
For this, DroneSeed has a special craft equipped with a sophisticated imaging stack. This first pass is done using waypoints set on satellite imagery.
The information collected at this point is really far more detailed than what’s actually needed. The lidar, for instance, collects spatial information at a resolution much beyond what’s needed to understand the shape of the terrain and major obstacles. It produces a 3D map of the vegetation as well as the terrain, allowing the system to identify stumps, roots, bushes, new trees, erosion and other important features.
This works hand in hand with the multispectral camera, which collects imagery not just in the visible bands — useful for identifying things — but also in those outside the human range, which allows for in-depth analysis of the soil and plant life.
The resulting map of the area is not just useful for drone navigation, but for the surgical strikes that are necessary to make this kind of drone-based operation worth doing in the first place. No doubt there are researchers who would love to have this data as well.
Now, spraying and planting are very different tasks. The first tends to be done indiscriminately using helicopters, and the second by laborers who burn out after a couple of years — as mentioned above, it’s incredibly difficult work. The challenge in the first case is to improve efficiency and efficacy, while in the second case is to automate something that requires considerable intelligence.
Spraying is in many ways simpler. Identifying invasive plants isn’t easy, exactly, but it can be done with imagery like that the drones are collecting. Having identified patches of a plant to be eliminated, the drones can calculate a path and expend only as much herbicide is necessary to kill them, instead of dumping hundreds of gallons indiscriminately on the entire area. It’s cheaper and more environmentally friendly. Naturally, the opposite approach could be used for distributing fertilizer or some other agent.
I’m making it sound easy again. This isn’t a plug and play situation — you can’t buy a DJI drone and hit the “weedkiller” option in its control software. A big part of this operation was the creation not only of the drones themselves, but the infrastructure with which to deploy them.
Conservation convoy
The drones themselves are unique, but not alarmingly so. They’re heavy-duty craft, capable of lifting well over the 57 pounds of payload they carry (the FAA limits them to 115 pounds).
“We buy and gut aircraft, then retrofit them,” Canary explained simply. Their head of hardware, would probably like to think there’s a bit more to it than that, but really the problem they’re solving isn’t “make a drone” but “make drones plant trees.” To that end, Canary explained, “the most unique engineering challenge was building a planting module for the drone that functions with the software.” We’ll get to that later.
DroneSeed deploys drones in swarms, which means as many as five drones in the air at once — which in turn means they need two trucks and trailers with their boxes, power supplies, ground stations and so on. The company’s VP of operations comes from a military background where managing multiple aircraft onsite was part of the job, and she’s brought her rigorous command of multi-aircraft environments to the company.
The drones take off and fly autonomously, but always under direct observation by the crew. If anything goes wrong, they’re there to take over, though of course there are plenty of autonomous behaviors for what to do in case of, say, a lost positioning signal or bird strike.
They fly in patterns calculated ahead of time to be the most efficient, spraying at problem areas when they’re over them, and returning to the ground stations to have power supplies swapped out before returning to the pattern. It’s key to get this process down pat, since efficiency is a major selling point. If a helicopter does it in a day, why shouldn’t a drone swarm? It would be sad if they had to truck the craft back to a hangar and recharge them every hour or two. It also increases logistics costs like gas and lodging if it takes more time and driving.
This means the team involves several people, as well as several drones. Qualified pilots and observers are needed, as well as people familiar with the hardware and software that can maintain and troubleshoot on site — usually with no cell signal or other support. Like many other forms of automation, this one brings its own new job opportunities to the table.
AI plays Mother Nature
The actual planting process is deceptively complex.
The idea of loading up a drone with seeds and setting it free on a blasted landscape is easy enough to picture. Hell, it’s been done. There are efforts going back decades to essentially load seeds or seedlings into guns and fire them out into the landscape at speeds high enough to bury them in the dirt: in theory this combines the benefits of manual planting with the scale of carpeting the place with seeds.
But whether it was slapdash placement or the shock of being fired out of a seed gun, this approach never seemed to work.
Forestry researchers have shown the effectiveness of finding the right “microsite” for a seed or seedling; in fact, it’s why manual planting works as well as it does. Trained humans find perfect spots to put seedlings: in the lee of a log; near but not too near the edge of a stream; on the flattest part of a slope, and so on. If you really want a forest to grow, you need optimal placement, perfect conditions and preventative surgical strikes with pesticides.
Although it’s difficult, it’s also the kind of thing that a machine learning model can become good at. Sorting through messy, complex imagery and finding local minima and maxima is a specialty of today’s ML systems, and the aerial imagery from the drones is rich in relevant data.
The company’s CTO led the creation of an ML model that determines the best locations to put trees at a site — though this task can be highly variable depending on the needs of the forest. A logging company might want a tree every couple of feet, even if that means putting them in sub-optimal conditions — but a few inches to the left or right may make all the difference. On the other hand, national forests may want more sparse deployments or specific species in certain locations to curb erosion or establish sustainable firebreaks.
Once the data has been crunched, the map is loaded into the drones’ hive mind and the convoy goes to the location, where the craft are loaded with seeds instead of herbicides.
But not just any old seeds! You see, that’s one more wrinkle. If you just throw a sagebrush seed on the ground, even if it’s in the best spot in the world, it could easily be snatched up by an animal, roll or wash down to a nearby crevasse, or simply fail to find the right nutrients in time despite the planter’s best efforts.
That’s why DroneSeed’s head of Planting and his team have been working on a proprietary seed packet that they were unbelievably reticent to detail.
From what I could gather, they’ve put a ton of work into packaging the seeds into nutrient-packed little pucks held together with a biodegradable fiber. The outside is dusted with capsaicin, the chemical that makes spicy food spicy (and also what makes bear spray do what it does). If they hadn’t told me, I might have guessed, since the workshop area was hazy with it, leading us all to cough and tear up a little. If I were a marmot, I’d learn to avoid these things real fast.
The pucks, or “seed vessels,” can and must be customized for the location and purpose — you have to match the content and acidity of the soil, things like that. DroneSeed will have to make millions of these things, but it doesn’t plan to be the manufacturer.
Finally these pucks are loaded in a special puck-dispenser which, closely coordinating with the drone, spits one out at the exact moment and speed needed to put it within a few centimeters of the microsite.
All these factors should improve the survival rate of seedlings substantially. That means that the company’s methods will not only be more efficient, but more effective. Reforestation is a numbers game played at scale, and even slight improvements — and DroneSeed is promising more than that — are measured in square miles and millions of tons of biomass.
Proof of life
DroneSeed has already signed several big contracts for spraying, and planting is next. Unfortunately, the timing on their side meant they missed this year’s planting season, though by doing a few small sites and showing off the results, they’ll be in pole position for next year.
After demonstrating the effectiveness of the planting technique, the company expects to expand its business substantially. That’s the scaling part — again, not easy, but easier than hiring another couple thousand planters every year.
Ideally the hardware can be assigned to local teams that do the on-site work, producing loci of activity around major forests from which jobs can be deployed at large or small scales. A set of five or six drones does the work of one helicopter, roughly speaking, so depending on the volume requested by a company or forestry organization, you may need dozens on demand.
That’s all yet to be explored, but DroneSeed is confident that the industry will see the writing on the wall when it comes to the old methods, and identify them as a solution that fits the future.
If it sounds like I’m cheerleading for this company, that’s because I am. It’s not often in the world of tech startups that you find a group of people not just attempting to solve a serious problem — it’s common enough to find companies hitting this or that issue — but who have spent the time, gathered the expertise and really done the dirty, boots-on-the-ground work that needs to happen so it goes from great idea to real company.
That’s what I felt was the case with DroneSeed, and here’s hoping their work pays off — for their sake, sure, but mainly for ours.
Powered by WPeMatico
Limiting social media use reduced loneliness and depression in new experiment
The idea that social media can be harmful to our mental and emotional well-being is not a new one, but little has been done by researchers to directly measure the effect; surveys and correlative studies are at best suggestive. A new experimental study out of Penn State, however, directly links more social media use to worse emotional states, and less use to better.
To be clear on the terminology here, a simple survey might ask people to self-report that using Instagram makes them feel bad. A correlative study would, for example, find that people who report more social media use are more likely to also experience depression. An experimental study compares the results from an experimental group with their behavior systematically modified, and a control group that’s allowed to do whatever they want.
This study, led by Melissa Hunt at Penn State’s psychology department, is the latter — which despite intense interest in this field and phenomenon is quite rare. The researchers only identified two other experimental studies, both of which only addressed Facebook use.
One hundred and forty-three students from the school were monitored for three weeks after being assigned to either limit their social media use to about 10 minutes per app (Facebook, Snapchat and Instagram) per day or continue using it as they normally would. They were monitored for a baseline before the experimental period and assessed weekly on a variety of standard tests for depression, social support and so on. Social media usage was monitored via the iOS battery use screen, which shows app use.
The results are clear. As the paper, published in the latest Journal of Social and Clinical Psychology, puts it:
The limited use group showed significant reductions in loneliness and depression over three weeks compared to the control group. Both groups showed significant decreases in anxiety and fear of missing out over baseline, suggesting a benefit of increased self-monitoring.
Our findings strongly suggest that limiting social media use to approximately 30 minutes per day may lead to significant improvement in well-being.
It’s not the final word in this, however. Some scores did not see improvement, such as self-esteem and social support. And later follow-ups to see if feelings reverted or habit changes were less than temporary were limited because most of the subjects couldn’t be compelled to return. (Psychology, often summarized as “the study of undergraduates,” relies on student volunteers who have no reason to take part except for course credit, and once that’s given, they’re out.)
That said, it’s a straightforward causal link between limiting social media use and improving some aspects of emotional and social health. The exact nature of the link, however, is something at which Hunt could only speculate:
Some of the existing literature on social media suggests there’s an enormous amount of social comparison that happens. When you look at other people’s lives, particularly on Instagram, it’s easy to conclude that everyone else’s life is cooler or better than yours.
When you’re not busy getting sucked into clickbait social media, you’re actually spending more time on things that are more likely to make you feel better about your life.
The researchers acknowledge the limited nature of their study and suggest numerous directions for colleagues in the field to take it from here. A more diverse population, for instance, or including more social media platforms. Longer experimental times and comprehensive follow-ups well after the experiment would help, as well.
The 30-minute limit was chosen as a conveniently measurable one, but the team does not intend to say that it is by any means the “correct” amount. Perhaps half or twice as much time would yield similar or even better results, they suggest: “It may be that there is an optimal level of use (similar to a dose response curve) that could be determined.”
Until then, we can use common sense, Hunt suggested: “In general, I would say, put your phone down and be with the people in your life.”
Powered by WPeMatico
Watch biotech startups pitch at IndieBio’s demo day today
IndieBio, the biotech startup accelerator that’s produced heaps of notable companies (including several that have graced the Startup Battlefield), is holding its twice-annual demo day today at 3PM Pacific Time. An even dozen young companies will be pitching their work, from AI-informed research to artificial meat, and you can watch them present live right here.
The IndieBio program is a four-month one that takes companies at the seed stage, often researchers straight out of graduate programs or university research groups, and gets them into shape for a proper Silicon Valley debut. Right now the companies get $250K in funding to take part, as well as plenty of resources, which parent VC firm SOSV can surely afford these days, what with raising $150 million last year.
Off the top of my head I remember two companies that competed at Disrupt SF 2016, Amaryllis Nucleics and mFluiDx, both very technical and highly talented teams. I’m always rooting for these kinds of wet lab companies, and it sounds like the current batch has plenty.
Watch the live pitches starting at 3PM below, and consult the list below the video for a summary of the companies presenting. We’ll be watching too!
New Age Meats: Pig farms are hell on earth. New Age Meats is a “cell-based meat company” that’s looking to replace animal-based pork sausage with a cleaner, more ethical grown alternative that goes just as well with pancakes.
NovoNutrients: Another non-traditional protein source, NovoNutrients uses industrial CO2 emissions to produce high-protein bacteria, which are harvested and sold as sustainable feed stock for aquaculture animals like fish.
BioRosa: An early detection method for autism spectrum disorders using blood tests that could shift diagnosis time to well before the current four years of age to potentially before the child is born.
Chronus Health: Hospitals need to do blood tests constantly, but often have to send samples to a central lab, which can take hours or days. Chronus has made a portable device they claim can provide complete blood count and metabolic panel tests essentially in real time.
Clinicai: Colorectal cancer, like other cancers, is best treated when detected early — and collecting and analyzing stool samples is a big part of that. These guys made a (prototype) device that attaches to ordinary toilets and non-invasively does what it needs to do, which could help people worldwide get proactive diagnosis and care.
Convalesce: Parkinson’s is a stubborn and tragic disease, but Convalesce is working on a treatment method involving injecting stem cells directly into areas affected by neurodegeneration.
Oralta: You can floss, brush, and rinse, but bad news bacteria are still going to take up residence in your mouth. Oralta hopes to combat them with good bacteria, reinforced by probiotic supplements. Fight fire with fire!
Ember: If someone is having a heart attack and it’ll take the EMTs 8-12 minutes to arrive, but your neighbor is a nurse trained in CPR, wouldn’t it be nice if they could stop by and help? That’s the idea with Ember, which hopes to improve outcomes by connecting patients with health professionals nearby.
Filtricine: The cancer treatment method being pursued by this company, instead of adding something lethal to cancer cells into the bloodstream, subtracts what they need to live while leaving normal cells unharmed. It could combine effectiveness with a blessed lack of side effects to become another tool in oncologists’ arsenals.
Serenity Bioworks: Gene therapy is another important therapeutic tool for a variety of problems, but some viral delivery methods can be fought by the body as if it’s fighting infection. Serenity is working on a system that suppresses that immune response and allows the friendly virus to deliver its payload.
Quartolio: So much scientific literature is published every year that there’s no way doctors and researchers can keep up. Quartolio aims to apply national language processing to journal articles to find connections and research opportunities that might otherwise have gone unnoticed.
Stämm: Bioreactors are used in practically every branch of biotech, whether for testing or drug manufacturing. Stämm is advancing the art with a modular, scalable microfluidic platform with highly tunable physical and chemical parameters.
Powered by WPeMatico
Subterranean drone mapping startup Emesent raises $2.5M to autonomously delve the deep
Seemingly every industry is finding ways to use drones in some way or another, but deep underground it’s a different story. In the confines of a mine or pipeline, with no GPS and little or no light, off-the-shelf drones are helpless — but an Australian startup called Emesent is giving them the spatial awareness and intelligence to navigate and map those spaces autonomously.
Drones that work underground or in areas otherwise inaccessible by GPS and other common navigation techniques are being made possible by a confluence of technology and computing power, explained Emesent CEO and co-founder Stefan Hrabar. The work they would take over from people is the epitome of “dull, dirty, and dangerous” — the trifecta for automation.
The mining industry is undoubtedly the most interested in this sort of thing; mining is necessarily a very systematic process and one that involves repeated measurements of areas being blasted, cleared, and so on. Frequently these measurements must be made manually and painstakingly in dangerous circumstances.
One mining technique has ore being blasted from the vertical space between two tunnels; the resulting cavities, called “stopes,” have to be inspected regularly to watch for problems and note progress.
“The way they scan these stopes is pretty archaic,” said Hrabar. “These voids can be huge, like 40-50 meters horizontally. They have to go to the edge of this dangerous underground cliff and sort of poke this stick out into it and try to get a scan. It’s very sparse information and from only one point of view, there’s a lot of missing data.”
Emesent’s solution, Hovermap, involves equipping a standard DJI drone with a powerful lidar sensor and a powerful onboard computing rig that performs simultaneous location and mapping (SLAM) work fast enough that the craft can fly using it. You put it down near the stope and it takes off and does its thing.
“The surveyors aren’t at risk and the data is orders of magnitude better. Everything is running onboard the drone in real time for path planning — that’s our core IP,” Hrabar said. “The dev team’s background is in drone autonomy, collision avoidance, terrain following — basically the drone sensing its environment and doing the right thing.”
As you can see in the video below, the drone can pilot itself through horizontal tunnels (imagine cave systems or transportation infrastructure) or vertical ones (stopes and sinkholes), slowly working its way along and returning minutes later with the data necessary to build a highly detailed map. I don’t know about you, but if I could send a drone ahead into the inky darkness to check for pits and other scary features, I wouldn’t think twice.
The idea is to sell the whole stack to mining companies as a plug-and-play solution, but work on commercializing the SLAM software separately for those who want to license and customize it. A data play is also in the works, naturally:
“At the end of the day, mining companies don’t want a point cloud, they want a report. So it’s not just collecting the data but doing the analytics as well,” said Hrabar.
Emesent emerged from Data61, the tech arm of Commonwealth Scientific and Industrial Research Organisation, or CSIRO, an Australian agency not unlike our national lab system. Hrabar worked there for over a decade on various autonomy projects, and three years ago started on what would become this company, eventually passing through the agency’s “ON” internal business accelerator.
Data collected from a pass through a cave system.
“Just last week, actually, is when we left the building,” Hrabar noted. “We’ve raised the funding we need for 18 months of runway with no revenue. We really are already generating revenue, though.”
The $3.5 million (Australian) round comes largely from a new $200M CSIRO Innovation fund managed by Main Sequence Ventures. Hrabar suggested that another round might be warranted in a year or two when the company decides to scale and expand into other verticals.
DARPA will be making its own contribution after a fashion through its Subterranean Challenge, should (as seemly likely) Emesent achieve success in it (they’re already an approved participant). Hrabar was confident. “It’s pretty fortuitous,” he said. “We’ve been doing underground autonomy for years, and then DARPA announces this challenge on exactly what we’re doing.”
We’ll be covering the challenge and its participants separately. You can read more about Emesent at its website.
Powered by WPeMatico
Reef-rejuvenating LarvalBot spreads coral babies by the millions
The continuing die-off of the world’s coral reefs is a depressing reminder of the reality of climate change, but it’s also something we can actively push back on. Conservationists have a new tool to do so with LarvalBot, an underwater robot platform that may greatly accelerate efforts to re-seed old corals with healthy new polyps.
The robot has a history going back to 2015, when a prototype known as COTSbot was introduced, capable of autonomously finding and destroying the destructive crown of thorns starfish (hence the name). It has since been upgraded and revised by the team at the Queensland University of Technology, and in its hunter-killer form is known as the RangerBot.
But the same systems that let it safely navigate and monitor corals for invasive fauna also make it capable of helping these vanishing ecosystems more directly.
Great Barrier Reef coral spawn yearly in a mass event that sees the waters off north Queensland filled with eggs and sperm. Researchers at Southern Cross University have been studying how to reap this harvest and sow a new generation of corals. They collect the eggs and sperm and sequester them in floating enclosures, where they are given a week or so to develop into viable coral babies (not my term, but I like it). These coral babies are then transplanted carefully to endangered reefs.
LarvalBot comes into play in that last step.
“We aim to have two or three robots ready for the November spawn. One will carry about 200,000 larvae and the other about 1.2 million,” explained QUT’s Matthew Dunbabin in a news release. “During operation, the robots will follow preselected paths at constant altitude across the reef and a person monitoring will trigger the release of the larvae to maximise the efficiency of the dispersal.”
It’s something a diver would normally have to do, so the robot acts as a force multiplier — one that doesn’t require food or oxygen, as well. A few of these could do the work of dozens of rangers or volunteers.
“The surviving corals will start to grow and bud and form new colonies which will grow large enough after about three years to become sexually reproductive and complete the life cycle,” said Southern Cross’s Peter Harrison, who has been developing the larval restoration technique.
It’s not a quick fix by any means, but this artificial spreading of corals could vastly improve the chances of a given reef or area surviving the next few years and eventually becoming self-sufficient again.
Powered by WPeMatico