Tag Archives: level

#436176 We’re Making Progress in Explainable ...

Machine learning algorithms are starting to exceed human performance in many narrow and specific domains, such as image recognition and certain types of medical diagnoses. They’re also rapidly improving in more complex domains such as generating eerily human-like text. We increasingly rely on machine learning algorithms to make decisions on a wide range of topics, from what we collectively spend billions of hours watching to who gets the job.

But machine learning algorithms cannot explain the decisions they make.
How can we justify putting these systems in charge of decisions that affect people’s lives if we don’t understand how they’re arriving at those decisions?

This desire to get more than raw numbers from machine learning algorithms has led to a renewed focus on explainable AI: algorithms that can make a decision or take an action, and tell you the reasons behind it.

What Makes You Say That?
In some circumstances, you can see a road to explainable AI already. Take OpenAI’s GTP-2 model, or IBM’s Project Debater. Both of these generate text based on a large corpus of training data, and try to make it as relevant as possible to the prompt that’s given. If these models were also able to provide a quick run-down of the top few sources in that corpus of training data they were drawing information from, it may be easier to understand where the “argument” (or poetic essay about unicorns) was coming from.

This is similar to the approach Google is now looking at for its image classifiers. Many algorithms are more sensitive to textures and the relationship between adjacent pixels in an image, rather than recognizing objects by their outlines as humans do. This leads to strange results: some algorithms can happily identify a totally scrambled image of a polar bear, but not a polar bear silhouette.

Previous attempts to make image classifiers explainable relied on significance mapping. In this method, the algorithm would highlight the areas of the image that contributed the most statistical weight to making the decision. This is usually determined by changing groups of pixels in the image and seeing which contribute to the biggest change in the algorithm’s impression of what the image is. For example, if the algorithm is trying to recognize a stop sign, changing the background is unlikely to be as important as changing the sign.

Google’s new approach changes the way that its algorithm recognizes objects, by examining them at several different resolutions and searching for matches to different “sub-objects” within the main object. You or I might recognize an ambulance from its flashing lights, its tires, and its logo; we might zoom in on the basketball held by an NBA player to deduce their occupation, and so on. By linking the overall categorization of an image to these “concepts,” the algorithm can explain its decision: I categorized this as a cat because of its tail and whiskers.

Even in this experiment, though, the “psychology” of the algorithm in decision-making is counter-intuitive. For example, in the basketball case, the most important factor in making the decision was actually the player’s jerseys rather than the basketball.

Can You Explain What You Don’t Understand?
While it may seem trivial, the conflict here is a fundamental one in approaches to artificial intelligence. Namely, how far can you get with mere statistical associations between huge sets of data, and how much do you need to introduce abstract concepts for real intelligence to arise?

At one end of the spectrum, Good Old-Fashioned AI or GOFAI dreamed up machines that would be entirely based on symbolic logic. The machine would be hard-coded with the concept of a dog, a flower, cars, and so forth, alongside all of the symbolic “rules” which we internalize, allowing us to distinguish between dogs, flowers, and cars. (You can imagine a similar approach to a conversational AI would teach it words and strict grammatical structures from the top down, rather than “learning” languages from statistical associations between letters and words in training data, as GPT-2 broadly does.)

Such a system would be able to explain itself, because it would deal in high-level, human-understandable concepts. The equation is closer to: “ball” + “stitches” + “white” = “baseball”, rather than a set of millions of numbers linking various pathways together. There are elements of GOFAI in Google’s new approach to explaining its image recognition: the new algorithm can recognize objects based on the sub-objects they contain. To do this, it requires at least a rudimentary understanding of what those sub-objects look like, and the rules that link objects to sub-objects, such as “cats have whiskers.”

The issue, of course, is the—maybe impossible—labor-intensive task of defining all these symbolic concepts and every conceivable rule that could possibly link them together by hand. The difficulty of creating systems like this, which could handle the “combinatorial explosion” present in reality, helped to lead to the first AI winter.

Meanwhile, neural networks rely on training themselves on vast sets of data. Without the “labeling” of supervised learning, this process might bear no relation to any concepts a human could understand (and therefore be utterly inexplicable).

Somewhere between these two, hope explainable AI enthusiasts, is a happy medium that can crunch colossal amounts of data, giving us all of the benefits that recent, neural-network AI has bestowed, while showing its working in terms that humans can understand.

Image Credit: Image by Seanbatty from Pixabay Continue reading

Posted in Human Robots

#436126 Quantum Computing Gets a Boost From AI ...

Illustration: Greg Mably

Anyone of a certain age who has even a passing interest in computers will remember the remarkable breakthrough that IBM made in 1997 when its Deep Blue chess-playing computer defeated Garry Kasparov, then the world chess champion. Computer scientists passed another such milestone in March 2016, when DeepMind (a subsidiary of Alphabet, Google’s parent company) announced that its AlphaGo program had defeated world-champion player Lee Sedol in the game of Go, a board game that had vexed AI researchers for decades. Recently, DeepMind’s algorithms have also bested human players in the computer games StarCraft IIand Quake Arena III.

Some believe that the cognitive capacities of machines will overtake those of human beings in many spheres within a few decades. Others are more cautious and point out that our inability to understand the source of our own cognitive powers presents a daunting hurdle. How can we make thinking machines if we don’t fully understand our own thought processes?

Citizen science, which enlists masses of people to tackle research problems, holds promise here, in no small part because it can be used effectively to explore the boundary between human and artificial intelligence.

Some citizen-science projects ask the public to collect data from their surroundings (as eButterfly does for butterflies) or to monitor delicate ecosystems (as Eye on the Reef does for Australia’s Great Barrier Reef). Other projects rely on online platforms on which people help to categorize obscure phenomena in the night sky (Zooniverse) or add to the understanding of the structure of proteins (Foldit). Typically, people can contribute to such projects without any prior knowledge of the subject. Their fundamental cognitive skills, like the ability to quickly recognize patterns, are sufficient.

In order to design and develop video games that can allow citizen scientists to tackle scientific problems in a variety of fields, professor and group leader Jacob Sherson founded ScienceAtHome (SAH), at Aarhus University, in Denmark. The group began by considering topics in quantum physics, but today SAH hosts games covering other areas of physics, math, psychology, cognitive science, and behavioral economics. We at SAH search for innovative solutions to real research challenges while providing insight into how people think, both alone and when working in groups.

It is computationally intractable to completely map out a higher-dimensional landscape: It is called the curse of high dimensionality, and it plagues many optimization problems.

We believe that the design of new AI algorithms would benefit greatly from a better understanding of how people solve problems. This surmise has led us to establish the Center for Hybrid Intelligence within SAH, which tries to combine human and artificial intelligence, taking advantage of the particular strengths of each. The center’s focus is on the gamification of scientific research problems and the development of interfaces that allow people to understand and work together with AI.

Our first game, Quantum Moves, was inspired by our group’s research into quantum computers. Such computers can in principle solve certain problems that would take a classical computer billions of years. Quantum computers could challenge current cryptographic protocols, aid in the design of new materials, and give insight into natural processes that require an exact solution of the equations of quantum mechanics—something normal computers are inherently bad at doing.

One candidate system for building such a computer would capture individual atoms by “freezing” them, as it were, in the interference pattern produced when a laser beam is reflected back on itself. The captured atoms can thus be organized like eggs in a carton, forming a periodic crystal of atoms and light. Using these atoms to perform quantum calculations requires that we use tightly focused laser beams, called optical tweezers, to transport the atoms from site to site in the light crystal. This is a tricky business because individual atoms do not behave like particles; instead, they resemble a wavelike liquid governed by the laws of quantum mechanics.

In Quantum Moves, a player manipulates a touch screen or mouse to move a simulated laser tweezer and pick up a trapped atom, represented by a liquidlike substance in a bowl. Then the player must bring the atom back to the tweezer’s initial position while trying to minimize the sloshing of the liquid. Such sloshing would increase the energy of the atom and ultimately introduce errors into the operations of the quantum computer. Therefore, at the end of a move, the liquid should be at a complete standstill.

To understand how people and computers might approach such a task differently, you need to know something about how computerized optimization algorithms work. The countless ways of moving a glass of water without spilling may be regarded as constituting a “solution landscape.” One solution is represented by a single point in that landscape, and the height of that point represents the quality of the solution—how smoothly and quickly the glass of water was moved. This landscape might resemble a mountain range, where the top of each mountain represents a local optimum and where the challenge is to find the highest peak in the range—the global optimum.

Illustration: Greg Mably

Researchers must compromise between searching the landscape for taller mountains (“exploration”) and climbing to the top of the nearest mountain (“exploitation”). Making such a trade-off may seem easy when exploring an actual physical landscape: Merely hike around a bit to get at least the general lay of the land before surveying in greater detail what seems to be the tallest peak. But because each possible way of changing the solution defines a new dimension, a realistic problem can have thousands of dimensions. It is computationally intractable to completely map out such a higher-dimensional landscape. We call this the curse of high dimensionality, and it plagues many optimization problems.

Although algorithms are wonderfully efficient at crawling to the top of a given mountain, finding good ways of searching through the broader landscape poses quite a challenge, one that is at the forefront of AI research into such control problems. The conventional approach is to come up with clever ways of reducing the search space, either through insights generated by researchers or with machine-learning algorithms trained on large data sets.

At SAH, we attacked certain quantum-optimization problems by turning them into a game. Our goal was not to show that people can beat computers in this arena but rather to understand the process of generating insights into such problems. We addressed two core questions: whether allowing players to explore the infinite space of possibilities will help them find good solutions and whether we can learn something by studying their behavior.

Today, more than 250,000 people have played Quantum Moves, and to our surprise, they did in fact search the space of possible moves differently from the algorithm we had put to the task. Specifically, we found that although players could not solve the optimization problem on their own, they were good at searching the broad landscape. The computer algorithms could then take those rough ideas and refine them.

Herbert A. Simon said that “solving a problem simply means representing it so as to make the solution transparent.” Apparently, that’s what our games can do with their novel user interfaces.

Perhaps even more interesting was our discovery that players had two distinct ways of solving the problem, each with a clear physical interpretation. One set of players started by placing the tweezer close to the atom while keeping a barrier between the atom trap and the tweezer. In classical physics, a barrier is an impenetrable obstacle, but because the atom liquid is a quantum-mechanical object, it can tunnel through the barrier into the tweezer, after which the player simply moved the tweezer to the target area. Another set of players moved the tweezer directly into the atom trap, picked up the atom liquid, and brought it back. We called these two strategies the “tunneling” and “shoveling” strategies, respectively.

Such clear strategies are extremely valuable because they are very difficult to obtain directly from an optimization algorithm. Involving humans in the optimization loop can thus help us gain insight into the underlying physical phenomena that are at play, knowledge that may then be transferred to other types of problems.

Quantum Moves raised several obvious issues. First, because generating an exceptional solution required further computer-based optimization, players were unable to get immediate feedback to help them improve their scores, and this often left them feeling frustrated. Second, we had tested this approach on only one scientific challenge with a clear classical analogue, that of the sloshing liquid. We wanted to know whether such gamification could be applied more generally, to a variety of scientific challenges that do not offer such immediately applicable visual analogies.

We address these two concerns in Quantum Moves 2. Here, the player first generates a number of candidate solutions by playing the original game. Then the player chooses which solutions to optimize using a built-in algorithm. As the algorithm improves a player’s solution, it modifies the solution path—the movement of the tweezer—to represent the optimized solution. Guided by this feedback, players can then improve their strategy, come up with a new solution, and iteratively feed it back into this process. This gameplay provides high-level heuristics and adds human intuition to the algorithm. The person and the machine work in tandem—a step toward true hybrid intelligence.

In parallel with the development of Quantum Moves 2, we also studied how people collaboratively solve complex problems. To that end, we opened our atomic physics laboratory to the general public—virtually. We let people from around the world dictate the experiments we would run to see if they would find ways to improve the results we were getting. What results? That’s a little tricky to explain, so we need to pause for a moment and provide a little background on the relevant physics.

One of the essential steps in building the quantum computer along the lines described above is to create the coldest state of matter in the universe, known as a Bose-Einstein condensate. Here millions of atoms oscillate in synchrony to form a wavelike substance, one of the largest purely quantum phenomena known. To create this ultracool state of matter, researchers typically use a combination of laser light and magnetic fields. There is no familiar physical analogy between such a strange state of matter and the phenomena of everyday life.

The result we were seeking in our lab was to create as much of this enigmatic substance as was possible given the equipment available. The sequence of steps to accomplish that was unknown. We hoped that gamification could help to solve this problem, even though it had no classical analogy to present to game players.

Images: ScienceAtHome

Fun and Games: The
Quantum Moves game evolved over time, from a relatively crude early version [top] to its current form [second from top] and then a major revision,
Quantum Moves 2 [third from top].
Skill Lab: Science Detective games [bottom] test players’ cognitive skills.

In October 2016, we released a game that, for two weeks, guided how we created Bose-Einstein condensates in our laboratory. By manipulating simple curves in the game interface, players generated experimental sequences for us to use in producing these condensates—and they did so without needing to know anything about the underlying physics. A player would generate such a solution, and a few minutes later we would run the sequence in our laboratory. The number of ultracold atoms in the resulting Bose-Einstein condensate was measured and fed back to the player as a score. Players could then decide either to try to improve their previous solution or to copy and modify other players’ solutions. About 600 people from all over the world participated, submitting 7,577 solutions in total. Many of them yielded bigger condensates than we had previously produced in the lab.

So this exercise succeeded in achieving our primary goal, but it also allowed us to learn something about human behavior. We learned, for example, that players behave differently based on where they sit on the leaderboard. High-performing players make small changes to their successful solutions (exploitation), while poorly performing players are willing to make more dramatic changes (exploration). As a collective, the players nicely balance exploration and exploitation. How they do so provides valuable inspiration to researchers trying to understand human problem solving in social science as well as to those designing new AI algorithms.

How could mere amateurs outperform experienced experimental physicists? The players certainly weren’t better at physics than the experts—but they could do better because of the way in which the problem was posed. By turning the research challenge into a game, we gave players the chance to explore solutions that had previously required complex programming to study. Indeed, even expert experimentalists improved their solutions dramatically by using this interface.

Insight into why that’s possible can probably be found in the words of the late economics Nobel laureate Herbert A. Simon: “Solving a problem simply means representing it so as to make the solution transparent [PDF].” Apparently, that’s what our games can do with their novel user interfaces. We believe that such interfaces might be a key to using human creativity to solve other complex research problems.

Eventually, we’d like to get a better understanding of why this kind of gamification works as well as it does. A first step would be to collect more data on what the players do while they are playing. But even with massive amounts of data, detecting the subtle patterns underlying human intuition is an overwhelming challenge. To advance, we need a deeper insight into the cognition of the individual players.

As a step forward toward this goal, ScienceAtHome created Skill Lab: Science Detective, a suite of minigames exploring visuospatial reasoning, response inhibition, reaction times, and other basic cognitive skills. Then we compare players’ performance in the games with how well these same people did on established psychological tests of those abilities. The point is to allow players to assess their own cognitive strengths and weaknesses while donating their data for further public research.

In the fall of 2018 we launched a prototype of this large-scale profiling in collaboration with the Danish Broadcasting Corp. Since then more than 20,000 people have participated, and in part because of the publicity granted by the public-service channel, participation has been very evenly distributed across ages and by gender. Such broad appeal is rare in social science, where the test population is typically drawn from a very narrow demographic, such as college students.

Never before has such a large academic experiment in human cognition been conducted. We expect to gain new insights into many things, among them how combinations of cognitive abilities sharpen or decline with age, what characteristics may be used to prescreen for mental illnesses, and how to optimize the building of teams in our work lives.

And so what started as a fun exercise in the weird world of quantum mechanics has now become an exercise in understanding the nuances of what makes us human. While we still seek to understand atoms, we can now aspire to understand people’s minds as well.

This article appears in the November 2019 print issue as “A Man-Machine Mind Meld for Quantum Computing.”

About the Authors
Ottó Elíasson, Carrie Weidner, Janet Rafner, and Shaeema Zaman Ahmed work with the ScienceAtHome project at Aarhus University in Denmark. Continue reading

Posted in Human Robots

#436100 Labrador Systems Developing Affordable ...

Developing robots for the home is still a challenge, especially if you want those robots to interact with people and help them do practical, useful things. However, the potential markets for home robots are huge, and one of the most compelling markets is for home robots that can assist humans who need them. Today, Labrador Systems, a startup based in California, is announcing a pre-seed funding round of $2 million (led by SOSV’s hardware accelerator HAX with participation from Amazon’s Alexa Fund and iRobot Ventures, among others) with the goal of expanding development and conducting pilot studies of “a new [assistive robot] platform for supporting home health.”

Labrador was founded two years ago by Mike Dooley and Nikolai Romanov. Both Mike and Nikolai have backgrounds in consumer robotics at Evolution Robotics and iRobot, but as an ’80s gamer, Mike’s bio (or at least the parts of his bio on LinkedIn) caught my attention: From 1995 to 1997, Mike worked at Brøderbund Software, helping to manage play testing for games like Myst and Riven and the Where in the World is Carmen San Diego series. He then spent three years at Lego as the product manager for MindStorms. After doing some marginally less interesting things, Mike was the VP of product development at Evolution Robotics from 2006 to 2012, where he led the team that developed the Mint floor sweeping robot. Evolution was acquired by iRobot in 2012, and Mike ended up as the VP of product development over there until 2017, when he co-founded Labrador.

I was pretty much sold at Where in the World is Carmen San Diego (the original version of which I played from a 5.25” floppy on my dad’s Apple IIe)*, but as you can see from all that other stuff, Mike knows what he’s doing in robotics as well.

And according to Labrador’s press release, what they’re doing is this:

Labrador Systems is an early stage technology company developing a new generation of assistive robots to help people live more independently. The company’s core focus is creating affordable solutions that address practical and physical needs at a fraction of the cost of commercial robots. … Labrador’s technology platform offers an affordable solution to improve the quality of care while promoting independence and successful aging.

Labrador’s personal robot, the company’s first offering, will enter pilot studies in 2020.

That’s about as light on detail as a press release gets, but there’s a bit more on Labrador’s website, including:

Our core focus is creating affordable solutions that address practical and physical needs. (we are not a social robot company)
By affordable, we mean products and technologies that will be available at less than 1/10th the cost of commercial robots.
We achieve those low costs by fusing the latest technologies coming out of augmented reality with robotics to move things in the real world.

The only hardware we’ve actually seen from Labrador at this point is a demo that they put together for Amazon’s re:MARS conference, which took place a few months ago, showing a “demonstration project” called Smart Walker:

This isn’t the home assistance robot that Labrador got its funding for, but rather a demonstration of some of their technology. So of course, the question is, what’s Labrador working on, then? It’s still a secret, but Mike Dooley was able to give us a few more details.

IEEE Spectrum: Your website shows a smart walker concept—how is that related to the assistive robot that you’re working on?

Mike Dooley: The smart walker was a request from a major senior living organization to have our robot (which is really good at navigation) guide residents from place to place within their communities. To test the idea with residents, it turned out to be much quicker to take the navigation system from the robot and put it on an existing rollator walker. So when you see the clips of the technology in the smart walker video on our website, that’s actually the robot’s navigation system localizing in real time and path planning in an environment.

“Assistive robot” can cover a huge range of designs and capabilities—can you give us any more detail about your robot, and what it’ll be able to do?

One of the core features of our robot is to help people move things where they have difficulty moving themselves, particularly in the home setting. That may sound trivial, but to someone who has impaired mobility, it can be a major daily challenge and negatively impact their life and health in a number of ways. Some examples we repeatedly hear are people not staying hydrated or taking their medication on time simply because there is a distance between where they are and the items they need. Once we have those base capabilities, i.e. the ability to navigate around a home and move things within it, then the robot becomes a platform for a wider variety of applications.

What made you decide to develop assistive robots, and why are robots a good solution for seniors who want to live independently?

Supporting independent living has been seen as a massive opportunity in robotics for some time, but also as something off in the future. The turning point for me was watching my mother enter that stage in her life and seeing her transition to using a cane, then a walker, and eventually to a wheelchair. That made the problems very real for me. It also made things much clearer about how we could start addressing specific needs with the tools that are becoming available now.

In terms of why robots can be a good solution, the basic answer is the level of need is so overwhelming that even helping with “basic” tasks can make an appreciable difference in the quality of someone’s daily life. It’s also very much about giving individuals a degree of control back over their environment. That applies to seniors as well as others whose world starts getting more complex to manage as their abilities become more impaired.

What are the particular challenges of developing assistive robots, and how are you addressing them? Why do you think there aren’t more robotics startups in this space?

The setting (operating in homes and personal spaces) and the core purpose of the product (aiding a wide variety of individuals) bring a lot of complexity to any capability you want to build into an assistive robot. Our approach is to put as much structure as we can into the system to make it functional, affordable, understandable and reliable.

I think one of the reasons you don’t see more startups in the space is that a lot of roboticists want to skip ahead and do the fancy stuff, such as taking on human-level capabilities around things like manipulation. Those are very interesting research topics, but we think those are also very far away from being practical solutions you can productize for people to use in their homes.

How do you think assistive robots and human caregivers should work together?

The ideal scenario is allowing caregivers to focus more of their time on the high-touch, personal side of care. The robot can offload the more basic support tasks as well as extend the impact of the caregiver for the long hours of the day they can’t be with someone at their home. We see that applying to both paid care providers as well as the 40 million unpaid family members and friends that provide assistance.

The robot is really there as a tool, both for individuals in need and the people that help them. What’s promising in the research discussions we’ve had so far, is that even when a caregiver is present, giving control back to the individual for simple things can mean a lot in the relationship between them and the caregiver.

What should we look forward to from Labrador in 2020?

Our big goal in 2020 is to start placing the next version of the robot with individuals with different types of needs to let them experience it naturally in their own homes and provide feedback on what they like, what don’t like and how we can make it better. We are currently reaching out to companies in the healthcare and home health fields to participate in those studies and test specific applications related to their services. We plan to share more detail about those studies and the robot itself as we get further into 2020.

If you’re an organization (or individual) who wants to possibly try out Labrador’s prototype, the company encourages you to connect with them through their website. And as we learn more about what Labrador is up to, we’ll have updates for you, presumably in 2020.

[ Labrador Systems ]

* I just lost an hour of my life after finding out that you can play Where in the World is Carmen San Diego in your browser for free. Continue reading

Posted in Human Robots

#436065 From Mainframes to PCs: What Robot ...

This is a guest post. The views expressed here are solely those of the author and do not represent positions of IEEE Spectrum or the IEEE.

Autonomous robots are coming around slowly. We already got autonomous vacuum cleaners, autonomous lawn mowers, toys that bleep and blink, and (maybe) soon autonomous cars. Yet, generation after generation, we keep waiting for the robots that we all know from movies and TV shows. Instead, businesses seem to get farther and farther away from the robots that are able to do a large variety of tasks using general-purpose, human anatomy-inspired hardware.

Although these are the droids we have been looking for, anything that came close, such as Willow Garage’s PR2 or Rethink Robotics’ Baxter has bitten the dust. With building a robotic company being particularly hard, compounding business risk with technological risk, the trend goes from selling robots to selling actual services like mowing your lawn, provide taxi rides, fulfilling retail orders, or picking strawberries by the pound. Unfortunately for fans of R2-D2 and C-3PO, these kind of business models emphasize specialized, room- or fridge-sized hardware that is optimized for one very specific task, but does not contribute to a general-purpose robotic platform.

We have actually seen something very similar in the personal computer (PC) industry. In the 1950s, even though computers could be as big as an entire room and were only available to a selected few, the public already had a good idea of what computers would look like. A long list of fictional computers started to populate mainstream entertainment during that time. In a 1962 New York Times article titled “Pocket Computer to Replace Shopping List,” visionary scientist John Mauchly stated that “there is no reason to suppose the average boy or girl cannot be master of a personal computer.”

In 1968, Douglas Engelbart gave us the “mother of all demos,” browsing hypertext on a graphical screen and a mouse, and other ideas that have become standard only decades later. Now that we have finally seen all of this, it might be helpful to examine what actually enabled the computing revolution to learn where robotics is really at and what we need to do next.

The parallels between computers and robots

In the 1970s, mainframes were about to be replaced by the emerging class of mini-computers, fridge-sized devices that cost less than US $25,000 ($165,000 in 2019 dollars). These computers did not use punch-cards, but could be programmed in Fortran and BASIC, dramatically expanding the ease with which potential applications could be created. Yet it was still unclear whether mini-computers could ever replace big mainframes in applications that require fast and efficient processing of large amounts of data, let alone enter every living room. This is very similar to the robotics industry right now, where large-scale factory robots (mainframes) that have existed since the 1960s are seeing competition from a growing industry of collaborative robots that can safely work next to humans and can easily be installed and programmed (minicomputers). As in the ’70s, applications for these devices that reach system prices comparable to that of a luxury car are quite limited, and it is hard to see how they could ever become a consumer product.

Yet, as in the computer industry, successful architectures are quickly being cloned, driving prices down, and entirely new approaches on how to construct or program robotic arms are sprouting left and right. Arm makers are joined by manufacturers of autonomous carts, robotic grippers, and sensors. These components can be combined, paving the way for standard general purpose platforms that follow the model of the IBM PC, which built a capable, open architecture relying as much on commodity parts as possible.

General purpose robotic systems have not been successful for similar reasons that general purpose, also known as “personal,” computers took decades to emerge. Mainframes were custom-built for each application, while typewriters got smarter and smarter, not really leaving room for general purpose computers in between. Indeed, given the cost of hardware and the relatively little abilities of today’s autonomous robots, it is almost always smarter to build a special purpose machine than trying to make a collaborative mobile manipulator smart.

A current example is e-commerce grocery fulfillment. The current trend is to reserve underutilized parts of a brick-and-mortar store for a micro-fulfillment center that stores goods in little crates with an automated retrieval system and a (human) picker. A number of startups like Alert Innovation, Fabric, Ocado Technology, TakeOff Technologies, and Tompkins Robotics, to just name a few, have raised hundreds of millions of venture capital recently to build mainframe equivalents of robotic fulfillment centers. This is in contrast with a robotic picker, which would drive through the aisles to restock and pick from shelves. Such a robotic store clerk would come much closer to our vision of a general purpose robot, but would require many copies of itself that crowd the aisles to churn out hundreds of orders per hour as a microwarehouse could. Although eventually more efficient, the margins in retail are already low and make it unlikely that this industry will produce the technological jump that we need to get friendly C-3POs manning the aisles.

Startups have raised hundreds of millions of venture capital recently to build mainframe equivalents of robotic fulfillment centers. This is in contrast with a robotic picker, which would drive through the aisles to restock and pick from shelves, and would come much closer to our vision of a general purpose robot.

Mainframes were also attacked from the bottom. Fascination with the new digital technology has led to a hobbyist movement to create microcomputers that were sold via mail order or at RadioShack. Initially, a large number of small businesses was selling tens, at most hundreds, of devices, usually as a kit and with wooden enclosures. This trend culminated into the “1977 Trinity” in the form of the Apple II, the Commodore PET, and the Tandy TRS-80, complete computers that were sold for prices around $2500 (TRS) to $5000 (Apple) in today’s dollars. The main application of these computers was their programmability (in BASIC), which would enable consumers to “learn to chart your biorhythms, balance your checking account, or even control your home environment,” according to an original Apple advertisement. Similarly, there exists a myriad of gadgets that explore different aspects of robotics such as mobility, manipulation, and entertainment.

As in the fledgling personal computing industry, the advertised functionality was at best a model of the real deal. A now-famous milestone in entertainment robotics was the original Sony’s Aibo, a robotic dog that was advertised to have many properties that a real dog has such as develop its own personality, play with a toy, and interact with its owner. Released in 1999, and re-launched in 2018, the platform has a solid following among hobbyists and academics who like its programmability, but probably only very few users who accept the device as a pet stand-in.

There also exist countless “build-your-own-robotic-arm” kits. One of the more successful examples is the uArm, which sells for around $800, and is advertised to perform pick and place, assembly, 3D printing, laser engraving, and many other things that sound like high value applications. Using compelling videos of the robot actually doing these things in a constrained environment has led to two successful crowd-funding campaigns, and have established the robot as a successful educational tool.

Finally, there exist platforms that allow hobbyist programmers to explore mobility to construct robots that patrol your house, deliver items, or provide their users with telepresence abilities. An example of that is the Misty II. Much like with the original Apple II, there remains a disconnect between the price of the hardware and the fidelity of the applications that were available.

For computers, this disconnect began to disappear with the invention of the first electronic spreadsheet software VisiCalc that spun out of Harvard in 1979 and prompted many people to buy an entire microcomputer just to run the program. VisiCalc was soon joined by WordStar, a word processing application, that sold for close to $2000 in today’s dollars. WordStar, too, would entice many people to buy the entire hardware just to use the software. The two programs are early examples of what became known as “killer application.”

With factory automation being mature, and robots with the price tag of a minicomputer being capable of driving around and autonomously carrying out many manipulation tasks, the robotics industry is somewhere where the PC industry was between 1973—the release of the Xerox Alto, the first computer with a graphical user interface, mouse, and special software—and 1979—when microcomputers in the under $5000 category began to take off.

Killer apps for robots
So what would it take for robotics to continue to advance like computers did? The market itself already has done a good job distilling what the possible killer apps are. VCs and customers alike push companies who have set out with lofty goals to reduce their offering to a simple value proposition. As a result, companies that started at opposite ends often converge to mirror images of each other that offer very similar autonomous carts, (bin) picking, palletizing, depalletizing, or sorting solutions. Each of these companies usually serves a single application to a single vertical—for example bin-picking clothes, transporting warehouse goods, or picking strawberries by the pound. They are trying to prove that their specific technology works without spreading themselves too thin.

Very few of these companies have really taken off. One example is Kiva Systems, which turned into the logistic robotics division of Amazon. Kiva and others are structured around sound value propositions that are grounded in well-known user needs. As these solutions are very specialized, however, it is unlikely that they result into any economies of scale of the same magnitude that early computer users who bought both a spreadsheet and a word processor application for their expensive minicomputer could enjoy. What would make these robotic solutions more interesting is when functionality becomes stackable. Instead of just being able to do bin picking, palletizing, and transportation with the same hardware, these three skills could be combined to model entire processes.

A skill that is yet little addressed by startups and is historically owned by the mainframe equivalent of robotics is assembly of simple mechatronic devices. The ability to assemble mechatronic parts is equivalent to other tasks such as changing a light bulb, changing the batteries in a remote control, or tending machines like a lever-based espresso machine. These tasks would involve the autonomous execution of complete workflows possible using a single machine, eventually leading to an explosion of industrial productivity across all sectors. For example, picking up an item from a bin, arranging it on the robot, moving it elsewhere, and placing it into a shelf or a machine is a process that equally applies to a manufacturing environment, a retail store, or someone’s kitchen.

Image: Robotic Materials Inc.

Autonomous, vision and force-based assembly of the
Siemens robot learning challenge.

Even though many of the above applications are becoming possible, it is still very hard to get a platform off the ground without added components that provide “killer app” value of their own. Interesting examples are Rethink Robotics or the Robot Operating System (ROS). Rethink Robotics’ Baxter and Sawyer robots pioneered a great user experience (like the 1973 Xerox Alto, really the first PC), but its applications were difficult to extend beyond simple pick-and-place and palletizing and depalletizing items.

ROS pioneered interprocess communication software that was adapted to robotic needs (multiple computers, different programming languages) and the idea of software modularity in robotics, but—in the absence of a common hardware platform—hasn’t yet delivered a single application, e.g. for navigation, path planning, or grasping, that performs beyond research-grade demonstration level and won’t get discarded once developers turn to production systems. At the same time, an increasing number of robotic devices, such as robot arms or 3D perception systems that offer intelligent functionality, provide other ways to wire them together that do not require an intermediary computer, while keeping close control over the real-time aspects of their hardware.

Image: Robotic Materials Inc.

Robotic Materials GPR-1 combines a MIR-100 autonomous cart with an UR-5 collaborative robotic arm, an onRobot force/torque sensor and Robotic Materials’ SmartHand to perform out-of-the-box mobile assembly, bin picking, palletizing, and depalletizing tasks.

At my company, Robotic Materials Inc., we have made strides to identify a few applications such as bin picking and assembly, making them configurable with a single click by combining machine learning and optimization with an intuitive user interface. Here, users can define object classes and how to grasp them using a web browser, which then appear as first-class objects in a robot-specific graphical programming language. We have also done this for assembly, allowing users to stack perception-based picking and force-based assembly primitives by simply dragging and dropping appropriate commands together.

While such an approach might answer the question of a killer app for robots priced in the “minicomputer” range, it is unclear how killer app-type value can be generated with robots in the less-than-$5000 category. A possible answer is two-fold: First, with low-cost arms, mobility platforms, and entertainment devices continuously improving, a confluence of technology readiness and user innovation, like with the Apple II and VisiCalc, will eventually happen. For example, there is not much innovation needed to turn Misty into a home security system; the uArm into a low-cost bin-picking system; or an Aibo-like device into a therapeutic system for the elderly or children with autism.

Second, robots and their components have to become dramatically cheaper. Indeed, computers have seen an exponential reduction in price accompanied by an exponential increase in computational power, thanks in great part to Moore’s Law. This development has helped robotics too, allowing us to reach breakthroughs in mobility and manipulation due to the ability to process massive amounts of image and depth data in real-time, and we can expect it to continue to do so.

Is there a Moore’s Law for robots?
One might ask, however, how a similar dynamics might be possible for robots as a whole, including all their motors and gears, and what a “Moore’s Law” would look like for the robotics industry. Here, it helps to remember that the perpetuation of Moore’s Law is not the reason, but the result of the PC revolution. Indeed, the first killer apps for bookkeeping, editing, and gaming were so good that they unleashed tremendous consumer demand, beating the benchmark on what was thought to be physically possible over and over again. (I vividly remember 56 kbps to be the absolute maximum data rate for copper phone lines until DSL appeared.)

That these economies of scale are also applicable to mechatronics is impressively demonstrated by the car industry. A good example is the 2020 Prius Prime, a highly computerized plug-in hybrid, that is available for one third of the cost of my company’s GPR-1 mobile manipulator while being orders of magnitude more complex, sporting an electrical motor, a combustion engine, and a myriad of sensors and computers. It is therefore very well conceivable to produce a mobile manipulator that retails at one tenth of the cost of a modern car, once robotics enjoy similar mass-market appeal. Given that these robots are part of the equation, actively lowering cost of production, this might happen as fast as never before in the history of industrialization.

It is therefore very well conceivable to produce a mobile manipulator that retails at one tenth of the cost of a modern car, once robotics enjoy similar mass-market appeal.

There is one more driver that might make robots exponentially more capable: the cloud. Once a general purpose robot has learned or was programmed with a new skill, it could share it with every other robot. At some point, a grocer who buys a robot could assume that it already knows how to recognize and handle 99 percent of the retail items in the store. Likewise, a manufacturer can assume that the robot can handle and assemble every item available from McMaster-Carr and Misumi. Finally, families could expect a robot to know every kitchen item that Ikea and Pottery Barn is selling. Sounds like a labor intense problem, but probably more manageable than collecting footage for Google’s Street View using cars, tricycles, and snowmobiles, among other vehicles.

Strategies for robot startups
While we are waiting for these two trends—better and better applications and hardware with decreasing cost—to converge, we as a community have to keep exploring what the canonical robotic applications beyond mobility, bin picking, palletizing, depalletizing, and assembly are. We must also continue to solve the fundamental challenges that stand in the way of making these solutions truly general and robust.

For both questions, it might help to look at the strategies that have been critical in the development of the personal computer, which might equally well apply to robotics:

Start with a solution to a problem your customers have. Unfortunately, their problem is almost never that they need your sensor, widget, or piece of code, but something that already costs them money or negatively affects them in some other way. Example: There are many more people who had a problem calculating their taxes (and wanted to buy VisiCalc) than writing their own solution in BASIC.

Build as little of your own hardware as necessary. Your business model should be stronger than the margin you can make on the hardware. Why taking the risk? Example: Why build your own typewriter if you can write the best typewriting application that makes it worth buying a computer just for that?

If your goal is a platform, make sure it comes with a killer application, which alone justifies the platform cost. Example: Microcomputer companies came and went until the “1977 Trinity” intersected with the killer apps spreadsheet and word processors. Corollary: You can also get lucky.

Use an open architecture, which creates an ecosystem where others compete on creating better components and peripherals, while allowing others to integrate your solution into their vertical and stack it with other devices. Example: Both the Apple II and the IBM PC were completely open architectures, enabling many clones, thereby growing the user and developer base.

It’s worthwhile pursuing this. With most business processes already being digitized, general purpose robots will allow us to fill in gaps in mobility and manipulation, increasing productivity at levels only limited by the amount of resources and energy that are available, possibly creating a utopia in which creativity becomes the ultimate currency. Maybe we’ll even get R2-D2.

Nikolaus Correll is an associate professor of computer science at the University of Colorado at Boulder where he works on mobile manipulation and other robotics applications. He’s co-founder and CTO of Robotic Materials Inc., which is supported by the National Science Foundation and the National Institute of Standards and Technology via their Small Business Innovative Research (SBIR) programs. Continue reading

Posted in Human Robots

#436044 Want a Really Hard Machine Learning ...

What’s the world’s hardest machine learning problem? Autonomous vehicles? Robots that can walk? Cancer detection?

Nope, says Julian Sanchez. It’s agriculture.

Sanchez might be a little biased. He is the director of precision agriculture for John Deere, and is in charge of adding intelligence to traditional farm vehicles. But he does have a little perspective, having spent time working on software for both medical devices and air traffic control systems.

I met with Sanchez and Alexey Rostapshov, head of digital innovation at John Deere Labs, at the organization’s San Francisco offices last month. Labs launched in 2017 to take advantage of the area’s tech expertise, both to apply machine learning to in-house agricultural problems and to work with partners to build technologies that play nicely with Deere’s big green machines. Deere’s neighbors in San Francisco’s tech-heavy South of Market are LinkedIn, Salesforce, and Planet Labs, which puts it in a good position for recruiting.

“We’ve literally had folks knock on the door and say, ‘What are you doing here?’” says Rostapshov, and some return to drop off resumes.

Here’s why Sanchez believes agriculture is such a big challenge for artificial intelligence.

“It’s not just about driving tractors around,” he says, although autonomous driving technologies are part of the mix. (John Deere is doing a lot of work with precision GPS to improve autonomous driving, for example, and allow tractors to plan their own routes around fields.)

But more complex than the driving problem, says Sanchez, are the classification problems.

Corn: A Classic Classification Problem

Photo: Tekla Perry

One key effort, Sanchez says, are AI systems “that allow me to tell whether grain being harvested is good quality or low quality and to make automatic adjustment systems for the harvester.” The company is already selling an early version of this image analysis technology. But the many differences between grain types, and grains grown under different conditions, make this task a tough one for machine learning.

“Take corn,” Sanchez says. “Let’s say we are building a deep learning algorithm to detect this corn. And we take lots of pictures of kernels to give it. Say we pick those kernels in central Illinois. But, one mile over, the farmer planted a slightly different hybrid which has slightly different coloration of yellow. Meanwhile, this other farm harvested three days later in a field five miles away; it’s the same hybrid, but it also looks different.

“It’s an overwhelming classification challenge, and that’s just for corn. But you are not only doing it for corn, you have to add 20 more varieties of grain to the mix; and some, like canola, are almost microscopic.”

Even the ground conditions vary dramatically—far more than road conditions, Sanchez points out.

“Let’s say we are building a deep learning algorithm to detect how much residue is left on the soil after a harvest, including stubble and some chaff. Let’s drive 2,000 acres of fields in the Midwest looking at residue. That’s great, but I guarantee that if you go drive those the next year, it will look significantly different.

“Deep learning is great at interpolating conditions between what it knows; it is not good at extrapolating to situations it hasn’t seen. And in agriculture, you always feel that there is a set of conditions that you haven’t yet classified.”

A Flood of Big Data

The scale of the data is also daunting, Rostapshov points out. “We are one of the largest users of cloud computing services in the world,” he says. “We are gathering 5 to 15 million measurements per second from 130,000 connected machines globally. We have over 150 million acres in our databases, using petabytes and petabytes [of storage]. We process more data than Twitter does.”

Much of this information is so-called dirty data, that is, it doesn’t share the same format or structure, because it’s coming not only from a wide variety of John Deere machines, but also includes data from some 100 other companies that have access to the platform, including weather information, aerial imagery, and soil analyses.

As a result, says Sanchez, Deere has had to make “tremendous investments in back-end data cleanup.”

Deep learning is great at interpolating conditions between what it knows; it is not good at extrapolating to situations it hasn’t seen.”
—Julian Sanchez, John Deere

“We have gotten progressively more skilled at that problem,” he says. “We started simply by cleaning up our own data. You’d think it would be nice and neat, since it’s coming from our own machines, but there is a wide variety of different models and different years. Then we started geospatially tagging the agronomic data—the information about where you are applying herbicides and fertilizer and the like—coming in from our vehicles. When we started bringing in other data, from drones, say, we were already good at cleaning it up.”

John Deere’s Hiring Pitch

Hard problems can be a good thing to have for a company looking to hire machine learning engineers.

“Our opening line to potential recruits,” Sanchez says, “is ‘This stuff matters.’ Then, if we get a chance to talk to them more, we follow up with ‘Not only does this stuff matter, but the problems are really hard and interesting.’ When we explain the variability in farming and how we have to apply all the latest tools to these problems, we get their attention.”

Software engineers “know that feeding a growing population is a massive problem and are excited about the prospect of making a difference,” Rostapshov says.

Only 20 engineers work in the San Francisco labs right now, and that’s on a busy day—some of the researchers spend part of their time at Blue River Technology, a startup based in Sunnyvale that was acquired by Deere in 2017. About half of the researchers are focusing on AI. The Lab is in the process of doubling its office space (no word on staffing plans for that expansion yet).

“We are one of the largest users of cloud computing services in the world.”
—Alexey Rostapshov, John Deere Labs

Company-wide, Deere has thousands of software engineers, with many using AI and machine learning tools in their work, and about the same number of mechanical and electrical engineers, Sanchez reports. “If you look at our hiring 10 years ago,” he says, “it was heavily weighted to mechanical engineers. But if you look at those numbers now, it is by a large majority [engineers working] in the software space. We still need mechanical engineers—we do build green machines—but if you go by our footprint of tech talent, it is pretty safe to call John Deere a software company. And if you follow the key conversations that are happening in the company right now, 95 percent of them are software-related.”

For now, these software engineers are focused on developing technologies that allow farmers to “do more with less,” Sanchez says. Meaning, to get more and better crops from less fuel, less seed, less fertilizer, less pesticide, and fewer workers, and putting together building blocks that, he says, could eventually lead to fully autonomous farm vehicles. The data Deere collects today, for the most part, stays in silos (the virtual kind), with AI algorithms that analyze specific sets of data to provide guidance to individual farmers. At some point, however, with tools to anonymize data and buy-in from farmers, aggregating data could provide some powerful insights.

“We are not asking farmers for that yet,” Sanchez says. “We are not doing aggregation to look for patterns. We are focused on offering technology that allows an individual farmer to use less, on positioning ourselves to be in a neutral spot. We are not about selling you more seed or more fertilizer. So we are building up a good trust level. In the long term, we can have conversations about doing more with deep learning.” Continue reading

Posted in Human Robots