Collision course: a geological mystery in the Himalayas

According to Craig Martin, deciphering Earth’s geologic past is like an ant climbing over a car crash. “You’ve got to work out how the car crash happened, how fast the cars were going, at what angle they impacted,” explains Martin, a graduate student at MIT’s Department of Earth, Atmospheric and Planetary Sciences (EAPS). “You’re just a tiny ant wandering over this massive chaos,” he adds.

The crash site Martin is investigating is the Himalayas, a 1,400-mile mountain range that rose when the Indian and Eurasian tectonic plates scrunched together. “The mainstream idea is: There was Eurasia; there was India; and they collided 50 million years ago,” says Oliver Jagoutz, an associate professor in EAPS and Martin’s advisor. “We think it was much more complicated than that, because it’s always more complicated.”

Detective work at 11,000 feet

Eighty million years ago, India and Eurasia were 4,000 miles apart, separated by an ancient body of water that geologists call the Neotethys Ocean, but Jagoutz believes there was more than just seawater between the two. He’s not alone. Many geologists agree on the existence of an arc of volcanic islands that formed on the boundary of a smaller tectonic plate, similar to the Mariana Islands in the Pacific Ocean. However, there is debate on whether these islands first collided with the Eurasian plate to the north or the Indian plate to the south. Jagoutz’s hypothesis is the latter. “If I’m right, the arc sits near the equator. If the others are right, the fragments should be 30 degrees north,” he explains. “That’s how simple it is.” But it can mean a world of difference in terms of explaining the paleoclimate — not just in the Himalayas, but globally as well.

To confirm this idea, Jagoutz and Martin turned to paleomagnetism. Some rock minerals, such as magnetite, contain iron and orient themselves along Earth’s magnetic field. At the equator, magnetite in newly formed rocks will be parallel to the ground, but the further north or south it is, the more inclined it will be. “We can measure, essentially, the latitude that a rock was formed at,” explains Martin.

If you were to take a slice of the Kohistan-Ladakh region of the Himalayas in northern India, you would see a succession of rock layers representing the India plate and the Eurasia plate, with the volcanic island arc sandwiched in between. “That’s why Ladakh is a really cool place to go to, because you can walk though this whole collision,” says Martin.

In summer 2018, Martin and Jade Fischer, a junior double-majoring in EAPS and physics, spent six weeks in Ladakh collecting samples from the volcanic rocks. Back at MIT, Martin measured the paleomagnetic signature of these rocks, and his results placed the Kohistan-Ladakh arc right at the equator, in agreement with Jagoutz’s theory.

A magnetic collaboration

Megan Guenther, a junior in EAPS, first heard about the opportunity to do field work in Ladakh when Martin gave a presentation about his research in her structural geology class last fall. “At the end, he told us he was probably going again and to let him know if we were interested,” Guenther explains. “I emailed him an hour later.”

Guenther had been looking for a chance to gain more field experience. She works on the compositions of lunar glasses with Tim Grove, the Robert R. Shrock Professor of Earth and Planetary Sciences, where the research takes place entirely in the lab. “You can’t really do field work on the moon,” she jokes.

This past summer, Guenther and Martin spent six weeks in Ladakh collecting rock samples from the Eurasia plate to prove that this was not also further south, mapping the region and doing structural analyses. Both Guenther and Martin were supported by MIT International Science and Technology Initiatives (MISTI) and the MISTI Global Seed Fund.

MISTI and Jagoutz go back a long time, with MISTI funding class excursions, department field trips, and a number of Jagoutz’s students. “MISTI-India has been good to us,” he says. “They financed the workshop where we came up with the whole concept of this work.” And, says Jagoutz, the students really love the experience. “They get influenced by it, and a lot of people chose their career paths after it,” says Jagoutz. “Ultimately, that’s what MISTI is all about: an experience that tells students they want to get into science.”

For Guenther, the trip was an essential part of her education as a geologist. “I feel much more confident as a field geologist, which is exactly what I wanted,” she says. It also impressed on her the titanic scale of geology. “The scale of everything is so crazy,” says Guenther. “You’re already at 11,000 feet, minimum, the whole time, and then these huge mountains tower above that.”

By solving the story of the collision that resulted in the Himalayas, Jagoutz and his team also shed light on its global implications. Large-scale collisions, Jagoutz explains, don’t just have local effects, and in the case of the Himalayas they can also explain some of Earth’s past glaciation events. “That’s the good thing about geology: the dimensions,” says Jagoutz. “You look at a magnetite crystal in a rock, and it tells you how global cooling works.”

de MIT News https://ift.tt/2BO7TC1

3 Questions: Ariel White on the impact of incarceration on voting

Ariel White is an assistant professor of political science at MIT, where she researches voting and voting rights, race, the criminal legal system, and bureaucratic behavior. White uses large datasets (and sometimes experiments) to measure individual experiences and to shed light on people's everyday interactions with government. Her research has appeared in the American Political Science Review, Science, Political Behavior, and other journals.

Q: Your research has shown that even brief jail terms can depress voting. Given the United States’ extremely high incarceration rate, what are the implications of this finding for representative democracy in this country?

A: First, we have an enormous — and enormously racially disparate — system of arrest, conviction, and incarceration in this country. We incarcerate far, far more people than comparable countries. And then we also deny more people the vote than other countries, because some U.S. states prevent people from voting after felony convictions. As a result, many people are caught up in this system, and millions are legally prevented from voting even after they get released.

People of color, particularly African Americans, are especially likely to be legally barred from voting — partly because of higher conviction rates, and partly because they’re especially likely to live in states with strict disenfranchisement laws. This isn’t a coincidence; felon disenfranchisement laws in the South have a fairly explicit history of being targeted at African-Americans.

What’s more, my work suggests that felony disfranchisement (legal restrictions on voting by people convicted of felonies) is just the tip of the iceberg when it comes to voter participation. The people whose cases I studied were all facing misdemeanor, not felony, charges, which means that everyone sent to jail was still eligible to vote in the next election. These are people convicted of things like “stealing something worth less than $500,” which is not the sort of offense that we as a society have decided should result in disenfranchisement. Nevertheless, my research found spending time in jail was life-changing enough that these people became less likely to make it to the polls afterward.

Similarly, in states where people with felony convictions are allowed to vote, we still see them participating at really low rates. This means that there is an enormous number of people not voting, disproportionately concentrated in poor and minority neighborhoods. These neighborhoods may already struggle to get the attention of their elected representatives, and low voter turnout is not going to help the situation.

The story here is not just about whether people are legally allowed to vote, but about whether they are practically able to vote, whether they know they're allowed to vote, and whether they think they have any reason to do so. This is a question I'm working on right now: I’m trying to figure out whether there are ways to reincorporate people into the political system after they've had contact with the legal system.

Q: Are there proposals on the table that would allow more people to vote after having been incarcerated?

A: Yes. A number of states have been considering changes to their felony disenfranchisement policies to give more people the right to vote after a felony conviction, often by removing lifetime bans or by allowing people to vote while they are on probation or parole. Florida got a lot of attention last year when voters approved Amendment 4, which would allow people convicted of felonies to vote once they have completed their sentences. But it's not clear how many people have been re-enfranchised by this change — because after voters approved Amendment 4, the legislature decided that people won't be able to vote until they've paid any outstanding fines or fees they owe from their cases. With that barrier in place, it's hard to know how many people will actually qualify.

Also, felony disenfranchisement is just part of the story about how the criminal legal system affects democracy, as I’ve explained. I'm keeping a close eye on several other initiatives, including efforts to enable people to vote from jail. Many people serving time in jail have only been convicted of misdemeanor offenses; others may not have been convicted of anything at all, but are being held in jail while their cases proceed because they can't afford bail. In theory, people in both these categories have the right to vote. In practice, however, it can be very difficult to exercise that right.

Illinois recently passed a law to make it easier for people in these circumstances to vote, including setting up a temporary polling location inside the Cook County Jail. Some states have also introduced programs that inform people about their right to vote, such as when they are released from prison or complete parole. And, of course, broader criminal justice reform efforts that seek to reduce the use of jails and prisons could also have ripple effects for our electoral system.

Q: MIT President L. Rafael Reif has said that solving the great challenges of our time will require multidisciplinary problem-solving and “bilingual thinkers” — approaches that, from the beginning, bring together insights and expertise from the scientific, technological, and humanistic fields. In what ways do you think multidisciplinary collaboration can be used to improve American elections?

A: Multidisciplinary approaches can help us think both about frontline changes to election administration and about higher-level questions of voting policy. My colleague in political science, Professor Charles Stewart III, founded an amazing lab here at MIT — the MIT Election Data and Science Lab (informally, the MIT Election Lab) — that's a great example of this kind of collaboration.

The lab brings social scientists, computer scientists, and mathematicians together with election administrators and other practitioners, and they all dig into the nitty-gritty details of how to make elections run better. This can mean the ways we draw district maps, or how we choose voting machines, or how we tabulate votes and audit counts to make sure they're accurate. I think that's a great model for cross-disciplinary work.

There's also a broader lesson here for multidisciplinary work: While academics have useful tools and data, it's critically important to hear from people with personal knowledge of voting policies. Many of the proposals we've seen lately for making voting more accessible or re-enfranchising people have come from previously incarcerated people. So I think one important way to expand multidisciplinary thinking is for academic researchers to collaborate with people with a firsthand understanding of how policies play out on the ground.

 
Story prepared by MIT SHASS Commiunications
Editorial Team: Emily Hiestand and Kathryn O'Neill

de MIT News https://ift.tt/2pohv3R

Nature can help solve optimization problems

Today's best digital computers still struggle to solve, in a practical time frame, a certain class of problem: combinatorial optimization problems, or those that involve combing through large sets of possibilities to find the best solution. Quantum computers hold potential to take on these problems, but scaling up the number of quantum bits in these systems remains a hurdle. 

Now, MIT Lincoln Laboratory researchers have demonstrated an alternative, analog-based way to accelerate the computing of these problems. "Our computer works by 'computing with physics' and uses nature itself to help solve these tough optimization problems," says Jeffrey Chou, co-lead author of a paper about this work published in Nature's Scientific Reports. "It's made of standard electronic components, allowing us to scale our computer quickly and cheaply by leveraging the existing microchip industry."

Perhaps the most well-known combinatorial optimization problem is that of the traveling salesperson. The problem asks to find the shortest route a salesperson can take through a number of cities, starting and ending at the same one. It may seem simple with only a few cities, but the problem becomes exponentially difficult to solve as the number of cities grows, bogging down even the best supercomputers. Yet optimization problems need to be solved in the real world daily; the solutions are used to schedule shifts, minimize financial risk, discover drugs, plan shipments, reduce interference on wireless networks, and much more.

"It has been known for a very long time that digital computers are fundamentally bad at solving these types of problems," says Suraj Bramhavar, also a co-lead author. "Many of the algorithms that have been devised to find solutions have to trade off solution quality for time. Finding the absolute optimum solution winds up taking an unreasonably long time when the problem sizes grow." Finding better solutions and doing so in dramatically less time could save industries billions of dollars. Thus, researchers have been searching for new ways to build systems designed specifically for optimization.

Finding the beat  

Nature likes to optimize energy, or achieve goals in the most efficient and distributed manner. This principle can be witnessed in the synchrony of nature, like heart cells beating together or schools of fish moving as one. Similarly, if you set two pendulum clocks on the same surface, no matter when the individual pendula are set into motion, they will eventually be lulled into a synchronized rhythm, reaching their apex at the same time but moving in opposite directions (or out of phase). This phenomenon was first observed in 1665 by the Dutch scientist Christiaan Huygens. These clocks are an example of coupled oscillators, set up in such a way that energy can be transferred between them. 

"We've essentially built an electronic, programmable version of this [clock setup] using coupled nonlinear oscillators," Chou says, showing a YouTube video of metronomes displaying a similar phenomenon. "The idea is that if you set up a system that encodes your problem's energy landscape, then the system will naturally try to minimize the energy by synchronizing, and in doing so, will settle on the best solution. We can then read out this solution."

The laboratory's prototype is a type of Ising machine, a computer based on a model in physics that describes a network of magnets, each of which have a magnetic "spin" orientation that can point only up or down. Each spin's final orientation depends on its interaction with every other spin. The individual spin-to-spin interactions are defined with a specific coupling weight, which denotes the strength of their connection. The goal of an Ising machine is to find, given a specific coupling strength network, the correct configuration of each spin, up or down, that minimizes the overall system energy.

But how does an Ising machine solve an optimization problem? It turns out that optimization problems can be mapped directly onto the Ising model, so that a set of a spins with certain coupling weights can represent each city and the distances between them in the traveling salesperson problem. Thus, finding the lowest-energy configuration of spins in the Ising model translates directly into the solution for the seller's fastest route. However, solving this problem by individually checking each of the possible configurations becomes prohibitively difficult when the problems grow to even modest sizes. 

In recent years, there have been efforts to build quantum machines that map to the Ising model, the most notable of which is one from the Canadian company D-Wave Systems. These machines may offer an efficient way to search the large solution space and find the correct answer, although they operate at cryogenic temperatures.

The laboratory's system runs a similar search, but does so using simple electronic oscillators. Each oscillator represents a spin in the Ising model, and similarly takes on a binarized phase, where oscillators that are synchronized, or in phase, represent the "spin up" configuration and those that are out of phase represent the "spin down" configuration. To set the system up to solve an optimization problem, the problem is first mapped to the Ising model, translating it into programmable coupling weights connecting each oscillator.

With the coupling weights programmed, the oscillators are allowed to run, like the pendulum arm of each clock being released. The system then naturally relaxes to its overall minimum energy state. Electronically reading out each oscillator's final phase, representing "spin up" or "spin down," presents the answer to the posed question. When the system ran against more than 2,000 random optimization problems, it came to the correct solution 98 percent of the time.

Previously, researchers at Stanford University demonstrated an Ising machine that uses lasers and electronics to solve optimization problems. That work revealed the potential for a significant speedup over digital computing although, according to Chou, the system may be difficult and costly to scale to larger sizes. The goal of finding a simpler alternative ignited the laboratory's research. 

Scaling up

The individual oscillator circuit the team used in their demonstration is similar to circuitry found inside cellphones or Wi-Fi routers. One addition they've made is a crossbar architecture that allows all of the oscillators in the circuit to be directly coupled to each other. "We have found an architecture that is both scalable to manufacture and can enable full connectivity to thousands of oscillators," Chou says. A fully connected system allows it to easily be mapped to a wide variety of optimization problems. 

"This work from Lincoln Laboratory makes innovative use of a crossbar architecture in its construction of an analog-electronic Ising machine," says Peter McMahon, an assistant professor of applied and engineering physics at Cornell University who was not involved in this research. "It will be interesting to see how future developments of this architecture and platform perform."  

The laboratory's prototype Ising machine uses four oscillators. The team is now working out a plan to scale the prototype to larger numbers of oscillators, or "nodes," and fabricate it on a printed circuit board. "If we can get to, say, 500 nodes, there is a chance we can start to compete with existing computers, and at 1,000 nodes we might be able to beat them," Bramhavar says.

The team sees a clear path forward to scaling up because the technology is based on standard electronic components. It's also extremely cheap. All the parts for their prototype can be found in a typical undergraduate electrical engineering lab and were bought online for about $20.

"What excites me is the simplicity," Bramhavar adds. "Quantum computers are expected to demonstrate amazing performance, but the scientific and engineering challenges required to scale them up are quite hard. Demonstrating even a small fraction of the performance gains envisioned with quantum computers, but doing so using hardware from the existing electronics industry, would be a huge leap forward. Exploiting the natural behavior of these circuits to solve real problems presents a very compelling alternative for what the next era of computing could be."

de MIT News https://ift.tt/2NcTfJG

Supercomputer analyzes web traffic across entire internet

Using a supercomputing system, MIT researchers have developed a model that captures what web traffic looks like around the world on a given day, which can be used as a measurement tool for internet research and many other applications.

Understanding web traffic patterns at such a large scale, the researchers say, is useful for informing internet policy, identifying and preventing outages, defending against cyberattacks, and designing more efficient computing infrastructure. A paper describing the approach was presented at the recent IEEE High Performance Extreme Computing Conference.

For their work, the researchers gathered the largest publicly available internet traffic dataset, comprising 50 billion data packets exchanged in different locations across the globe over a period of several years.

They ran the data through a novel “neural network” pipeline operating across 10,000 processors of the MIT SuperCloud, a system that combines computing resources from the MIT Lincoln Laboratory and across the Institute. That pipeline automatically trained a model that captures the relationship for all links in the dataset — from common pings to giants like Google and Facebook, to rare links that only briefly connect yet seem to have some impact on web traffic.  

The model can take any massive network dataset and generate some statistical measurements about how all connections in the network affect each other. That can be used to reveal insights about peer-to-peer filesharing, nefarious IP addresses and spamming behavior, the distribution of attacks in critical sectors, and traffic bottlenecks to better allocate computing resources and keep data flowing.

In concept, the work is similar to measuring the cosmic microwave background of space, the near-uniform radio waves traveling around our universe that have been an important source of information to study phenomena in outer space. “We built an accurate model for measuring the background of the virtual universe of the Internet,” says Jeremy Kepner, a researcher at the MIT Lincoln Laboratory Supercomputing Center and an astronomer by training. “If you want to detect any variance or anomalies, you have to have a good model of the background.”

Joining Kepner on the paper are: Kenjiro Cho of the Internet Initiative Japan; KC Claffy of the Center for Applied Internet Data Analysis at the University of California at San Diego; Vijay Gadepally and Peter Michaleas of Lincoln Laboratory’s Supercomputing Center; and Lauren Milechin, a researcher in MIT’s Department of Earth, Atmospheric and Planetary Sciences.

Breaking up data

In internet research, experts study anomalies in web traffic that may indicate, for instance, cyber threats. To do so, it helps to first understand what normal traffic looks like. But capturing that has remained challenging. Traditional “traffic-analysis” models can only analyze small samples of data packets exchanged between sources and destinations limited by location. That reduces the model’s accuracy.

The researchers weren’t specifically looking to tackle this traffic-analysis issue. But they had been developing new techniques that could be used on the MIT SuperCloud to process massive network matrices. Internet traffic was the perfect test case.

Networks are usually studied in the form of graphs, with actors represented by nodes, and links representing connections between the nodes. With internet traffic, the nodes vary in sizes and location. Large supernodes are popular hubs, such as Google or Facebook. Leaf nodes spread out from that supernode and have multiple connections to each other and the supernode. Located outside that “core” of supernodes and leaf nodes are isolated nodes and links, which connect to each other only rarely.

Capturing the full extent of those graphs is infeasible for traditional models. “You can’t touch that data without access to a supercomputer,” Kepner says.

In partnership with the Widely Integrated Distributed Environment (WIDE) project, founded by several Japanese universities, and the Center for Applied Internet Data Analysis (CAIDA), in California, the MIT researchers captured the world’s largest packet-capture dataset for internet traffic. The anonymized dataset contains nearly 50 billion unique source and destination data points between consumers and various apps and services during random days across various locations over Japan and the U.S., dating back to 2015.

Before they could train any model on that data, they needed to do some extensive preprocessing. To do so, they utilized software they created previously, called Dynamic Distributed Dimensional Data Mode (D4M), which uses some averaging techniques to efficiently compute and sort “hypersparse data” that contains far more empty space than data points. The researchers broke the data into units of about 100,000 packets across 10,000 MIT SuperCloud processors. This generated more compact matrices of billions of rows and columns of interactions between sources and destinations.

Capturing outliers

But the vast majority of cells in this hypersparse dataset were still empty. To process the matrices, the team ran a neural network on the same 10,000 cores. Behind the scenes, a trial-and-error technique started fitting models to the entirety of the data, creating a probability distribution of potentially accurate models.

Then, it used a modified error-correction technique to further refine the parameters of each model to capture as much data as possible. Traditionally, error-correcting techniques in machine learning will try to reduce the significance of any outlying data in order to make the model fit a normal probability distribution, which makes it more accurate overall. But the researchers used some math tricks to ensure the model still saw all outlying data — such as isolated links — as significant to the overall measurements.

In the end, the neural network essentially generates a simple model, with only two parameters, that describes the internet traffic dataset, “from really popular nodes to isolated nodes, and the complete spectrum of everything in between,” Kepner says.

The researchers are now reaching out to the scientific community to find their next application for the model. Experts, for instance, could examine the significance of the isolated links the researchers found in their experiments that are rare but seem to impact web traffic in the core nodes.

Beyond the internet, the neural network pipeline can be used to analyze any hypersparse network, such as biological and social networks. “We’ve now given the scientific community a fantastic tool for people who want to build more robust networks or detect anomalies of networks,” Kepner says. “Those anomalies can be just normal behaviors of what users do, or it could be people doing things you don’t want.”

de MIT News https://ift.tt/2Whnp2C

Helping autonomous vehicles see around corners

To improve the safety of autonomous systems, MIT engineers have developed a system that can sense tiny changes in shadows on the ground to determine if there’s a moving object coming around the corner.  

Autonomous cars could one day use the system to quickly avoid a potential collision with another car or pedestrian emerging from around a building’s corner or from in between parked cars. In the future, robots that may navigate hospital hallways to make medication or supply deliveries could use the system to avoid hitting people.

In a paper being presented at next week’s International Conference on Intelligent Robots and Systems (IROS), the researchers describe successful experiments with an autonomous car driving around a parking garage and an autonomous wheelchair navigating hallways. When sensing and stopping for an approaching vehicle, the car-based system beats traditional LiDAR — which can only detect visible objects — by more than half a second.

That may not seem like much, but fractions of a second matter when it comes to fast-moving autonomous vehicles, the researchers say.

“For applications where robots are moving around environments with other moving objects or people, our method can give the robot an early warning that somebody is coming around the corner, so the vehicle can slow down, adapt its path, and prepare in advance to avoid a collision,” adds co-author Daniela Rus, director of the Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Andrew and Erna Viterbi Professor of Electrical Engineering and Computer Science. “The big dream is to provide ‘X-ray vision’ of sorts to vehicles moving fast on the streets.”

Currently, the system has only been tested in indoor settings. Robotic speeds are much lower indoors, and lighting conditions are more consistent, making it easier for the system to sense and analyze shadows.

Joining Rus on the paper are: first author Felix Naser SM ’19, a former CSAIL researcher; Alexander Amini, a CSAIL graduate student; Igor Gilitschenski, a CSAIL postdoc; recent graduate Christina Liao ’19; Guy Rosman of the Toyota Research Institute; and Sertac Karaman, an associate professor of aeronautics and astronautics at MIT.

Extending ShadowCam

For their work, the researchers built on their system, called “ShadowCam,” that uses computer-vision techniques to detect and classify changes to shadows on the ground. MIT professors William Freeman and Antonio Torralba, who are not co-authors on the IROS paper, collaborated on the earlier versions of the system, which were presented at conferences in 2017 and 2018.

For input, ShadowCam uses sequences of video frames from a camera targeting a specific area, such as the floor in front of a corner. It detects changes in light intensity over time, from image to image, that may indicate something moving away or coming closer. Some of those changes may be difficult to detect or invisible to the naked eye, and can be determined by various properties of the object and environment. ShadowCam computes that information and classifies each image as containing a stationary object or a dynamic, moving one. If it gets to a dynamic image, it reacts accordingly.

Adapting ShadowCam for autonomous vehicles required a few advances. The early version, for instance, relied on lining an area with augmented reality labels called “AprilTags,” which resemble simplified QR codes. Robots scan AprilTags to detect and compute their precise 3D position and orientation relative to the tag. ShadowCam used the tags as features of the environment to zero in on specific patches of pixels that may contain shadows. But modifying real-world environments with AprilTags is not practical.

The researchers developed a novel process that combines image registration and a new visual-odometry technique. Often used in computer vision, image registration essentially overlays multiple images to reveal variations in the images. Medical image registration, for instance, overlaps medical scans to compare and analyze anatomical differences.

Visual odometry, used for Mars Rovers, estimates the motion of a camera in real-time by analyzing pose and geometry in sequences of images. The researchers specifically employ “Direct Sparse Odometry” (DSO), which can compute feature points in environments similar to those captured by AprilTags. Essentially, DSO plots features of an environment on a 3D point cloud, and then a computer-vision pipeline selects only the features located in a region of interest, such as the floor near a corner. (Regions of interest were annotated manually beforehand.)

As ShadowCam takes input image sequences of a region of interest, it uses the DSO-image-registration method to overlay all the images from same viewpoint of the robot. Even as a robot is moving, it’s able to zero in on the exact same patch of pixels where a shadow is located to help it detect any subtle deviations between images.

Next is signal amplification, a technique introduced in the first paper. Pixels that may contain shadows get a boost in color that reduces the signal-to-noise ratio. This makes extremely weak signals from shadow changes far more detectable. If the boosted signal reaches a certain threshold — based partly on how much it deviates from other nearby shadows — ShadowCam classifies the image as “dynamic.” Depending on the strength of that signal, the system may tell the robot to slow down or stop.

“By detecting that signal, you can then be careful. It may be a shadow of some person running from behind the corner or a parked car, so the autonomous car can slow down or stop completely,” Naser says.

Tag-free testing

In one test, the researchers evaluated the system’s performance in classifying moving or stationary objects using AprilTags and the new DSO-based method. An autonomous wheelchair steered toward various hallway corners while humans turned the corner into the wheelchair’s path. Both methods achieved the same 70-percent classification accuracy, indicating AprilTags are no longer needed.

In a separate test, the researchers implemented ShadowCam in an autonomous car in a parking garage, where the headlights were turned off, mimicking nighttime driving conditions. They compared car-detection times versus LiDAR. In an example scenario, ShadowCam detected the car turning around pillars about 0.72 seconds faster than LiDAR. Moreover, because the researchers had tuned ShadowCam specifically to the garage’s lighting conditions, the system achieved a classification accuracy of around 86 percent.

Next, the researchers are developing the system further to work in different indoor and outdoor lighting conditions. In the future, there could also be ways to speed up the system’s shadow detection and automate the process of annotating targeted areas for shadow sensing.

This work was funded by the Toyota Research Institute.

de MIT News https://ift.tt/364TNdv

J-PAL North America announces first partners through Work of the Future Innovation Competition

J-PAL North America, a research center at MIT, will partner with four organizations to test promising models to help workers navigate the shifting labor market as part of the center’s inaugural Work of the Future Innovation Competition. 

Currently in its first year, J-PAL North America’s Work of the Future Initiative provides direct support to organizations and agencies interested in evaluating programs or policies related to the changing nature of work in North America. In the coming year, J-PAL North America will partner with the Center for Work Education and Employment (CWEE), Checkr, the City of Los Angeles Mayor's Innovation Team, and the Montana Department of Labor and Industry to develop and support rigorous evaluations of programs seeking to improve outcomes for workers. These programs aim to reduce barriers to employment, support workers as they navigate the complex job market, and bolster jobseekers’ abilities to secure and retain quality work. 

"It's exciting to see so many promising proposals in just the first year of the Work of the Future Initiative," says Lawrence Katz, co-scientific director of J-PAL North America and academic advisor to the Work of the Future Initiative. "We're hopeful that the initiative can continue to generate this level of enthusiasm as it seeks to develop promising programs and identify effective methods to help workers navigate the shifting labor market."   

CWEE is a workforce development agency based in Denver, Colorado, that provides low-income parents, the majority of whom are Temporary Assistance for Needy Families recipients, with complete career readiness and retention skills. CWEE will partner with J-PAL North America to develop an evaluation on the impact of its intensive case management and career readiness program on employment outcomes. 

“CWEE has been supporting low-income families in the Denver metro community for almost four decades, so it feels incumbent upon us to  take a deeper dive and better understand how the programs and support provided contribute to economic mobility,” says Katy Hamilton, CEO of CWEE. “It’s rare for a nonprofit of our size to have an opportunity like this to learn from the foremost thinkers in the space of academic assessments and social programs.”

Checkr is a background-check company focused on making hiring more inclusive and efficient. J-PAL North America will partner with Checkr to evaluate if the implementation of the Positive Adjudication Matrix (PAM) can reduce bias in the background-check and hiring process. PAM allows employers to deem certain types of offenses irrelevant to the roles for which they’re hiring. Companies can then choose to either filter out or de-emphasize these criminal records. 

“We believe advanced technology plays a critical role in reducing hiring bias,” says Checkr Co-Founder and CEO Daniel Yanisse. “With rigorous research, we can better develop products to ensure a fairer, more inclusive hiring process.”

The Mayor’s Innovation Team (i-team) works with Los Angeles, California, city departments to closely examine complex challenges and discover innovative solutions that can improve the quality of life in LA. The LA i-team will partner with J-PAL North America to design an evaluation of a program that helps job seekers better access employment services at the city's WorkSource Centers. The i-team seeks to apply behavioral science techniques and use technology to support job seekers, increase usage of the centers’ resources, and improve job placements that create opportunities for Angelenos and lift families out of poverty.

“Our mission is simple: We want to help people get access to resources that can change their lives,” says Los Angeles Mayor Eric Garcetti. “This partnership will help us expand that work — so that more Angelenos have opportunities to find a career, support their families, contribute to the economy, and strengthen our communities.” 

The Montana Department of Labor and Industry (MTDLI) is a government agency that promotes the well-being of Montana’s workers, employers, and citizens. MTDLI will partner with J-PAL North America to evaluate the effectiveness of Reemployment Services and Eligibility Assessments, a national program for individuals claiming unemployment insurance who have been identified as likely to exhaust their UI benefits.

“All organizations must set priorities and find the most effective solutions using limited resources, and MTDLI is no different. MTDLI strives to provide top-notch services to Montana’s workforce in cost-effective and successful ways,” says Barbara Wagner, chief economist at MTDLI. “Using data and research to drive decision-making helps us focus our resources on the best solutions, therefore allowing us to have a greater impact on our workers, businesses, and economy.”

In working with these four organizations J-PAL North America looks to support the development of randomized evaluations of strategies and innovations that address the changing nature of work in North America. Partnerships with the Center for Work Education and Employment, Checkr, the City of Los Angeles, and the Montana Department of Labor and Industry will help J-PAL North America generate solutions that increase opportunity, reduce disparities, and help all workers navigate shifts in the labor market.

J-PAL North America is a regional office of the Abdul Latif Jameel Poverty Action Lab. J-PAL was established in 2003 as a research center at MIT’s Department of Economics. Since then, it has built a global network of affiliated professors based at over 58 universities and regional offices in Africa, Europe, Latin America and the Caribbean, North America, South Asia, and Southeast Asia. J-PAL North America was established with support from the Alfred P. Sloan Foundation and Arnold Ventures and works to improve the effectiveness of social programs in North America through three core activities: research, policy outreach, and capacity building. The Work of the Future Initiative was launched with support from Arnold Ventures, the Bill and Melinda Gates Foundation, and the Gerri and Rich Wong Family.

de MIT News https://ift.tt/2pfbsyj

System prevents speedy drones from crashing in unfamiliar areas

Autonomous drones are cautious when navigating the unknown. They creep forward, frequently mapping unfamiliar areas before proceeding lest they crash into undetected objects. But this slowdown isn’t ideal for drones carrying out time-sensitive tasks, such as flying search-and-rescue missions through dense forests.  

Now MIT researchers have developed a trajectory-planning model that helps drones fly at high speeds through previously unexplored areas, while staying safe.

The model — aptly named “FASTER” — estimates the quickest possible path from a starting point to a destination point across all areas the drone can and can’t see, with no regard for safety. But, as the drone flies, the model continuously logs collision-free “back-up” paths that slightly deviate from that fast flight path. When the drone is unsure about a particular area, it detours down the back-up path and replans its path. The drone can thus cruise at high speeds along the quickest trajectory while occasionally slowing down slightly to ensure safety.

“We always want to execute the fastest path, but we don’t always know it’s safe. If, as we move along this fastest path, we discover there’s a problem, we need to have a backup plan,” says Jesus Tordesillas, a graduate student in the Department of Aeronautics and Astronautics (AeroAstro) and first author on a paper describing the model being presented at next month’s International Conference on Intelligent Robots and Systems. “We obtain a higher velocity trajectory that may not be safe and a slow-velocity trajectory that’s completely safe. The two paths are stitched together at first, but then one deviates for performance and the other for safety.”

In forest simulations, where a virtual drone navigates around cylinders representing trees, FASTER-powered drones safely completed flight paths about two times quicker than traditional models. In real-life tests, FASTER-powered drones maneuvering around cardboard boxes in a large room achieved speeds of 7.8 meters per second. That’s pushing limits for how fast the drones can fly, based on weight and reaction times, the researchers say.

“That’s about as fast as you can go,” says co-author Jonathan How, the Richard Cockburn Maclaurin Professor of Aeronautics and Astronautics. “If you were standing in a room with a drone flying 7 to 8 meters per second in it, you’d probably take a step back.[JPH1] ”

The paper’s other co-author is Brett T. Lopez, a former PhD student in AeroAstro and now a postdoc at NASA’s Jet Propulsion Laboratory.

Splitting paths

Drones use cameras to capture environment as voxels, 3D cubes generated from depth information. As the drone flies, each detected voxel gets labeled as “free-known space,” unoccupied by objects, and “occupied-known space,” which contains objects. The rest of the environment is “unknown space.” 

FASTER utilizes all of those areas to plan three types of trajectories — “whole,” “safe,” and “committed.” The whole trajectory is the entire path from starting point A to goal location B, through known and unknown areas. To do so, “convex decomposition,” a technique that breaks down complex models into discrete components, generates overlapping polyhedrons that model those three areas in an environment. Using some geometric techniques and mathematical constraints, the model uses these polyhedrons to compute an optimal whole trajectory.

Simultaneously, the model plans a safe trajectory. Somewhere along the whole trajectory, it plots a “rescue” point that indicates the last moment a drone can detour to unobstructed free-known space, based on its speed and other factors. To find a safe destination, it computes new polyhedrons that cover the free-known space. Then, it locates a spot inside these new polyhedrons. Basically, the drone stops in a spot that’s safe but as close as possible to unknown space, enabling a very quick and efficient detour.

Committed trajectory

The committed trajectory consists of the first interval of the whole trajectory, as well as the entire safe trajectory. But this first interval is independent of the safe trajectory, and therefore it is not affected by the braking needed for the safe trajectory.

The drone computes one whole trajectory at a time, while always keeping track of the safe trajectory. But it’s given a time limit: When it reaches the rescue point, it must have successfully computed the next whole trajectory through known or unknown space. If it does, it will continue following the whole trajectory. Otherwise, it diverts to the safe trajectory. This approach enables the drone to maintain high velocities along the committed trajectories, which is key to achieving high overall speeds.

For this to all work, the researchers designed ways for the drones to process all the planning data very quickly, which was challenging. Because the maps are so varied, for instance, the time limit given to each committed trajectory initially varied dramatically. That was computationally expensive and slowed down the drone’s planning, so the researchers developed a method to quickly compute fixed times for all the intervals along the trajectories, which simplified computations. The researchers also designed methods to reduce how many polyhedrons the drone must process to map its surroundings. Both of those methods dramatically increased planning times.

The researchers are currently building larger FASTER-powered drones with propellers designed to enable steady horizontal flight. Traditionally, drones will need to roll and pitch as they’re flying. But this custom drone would stay completely flat for various applications.

A potential application for FASTER, which has been developed with support by U.S. Department of Defense, could be improving search-and-rescue missions in forest environments, which present many planning and navigational challenges for autonomous drones. “But the unknown area doesn’t have to be forest,” How says. “It could be any area where you don’t know what’s coming, and it matters how quickly you acquire that knowledge. The main motivation is building more agile drones.”

de MIT News https://ift.tt/31FigCI