Tag Archives: machine learning

Stanford’s social robot ‘Jackrabbot’ seeks to understand pedestrian behavior

View video here.

The Computational Vision and Geometry Lab has developed a robot prototype that could soon autonomously move among us, following normal human social etiquettes. It’s named ‘Jackrabbot’ after the springy hares that bounce around campus.

BY VIGNESH RAMACHANDRAN


In order for robots to circulate on sidewalks and mingle with humans in other crowded places, they’ll have to understand the unwritten rules of pedestrian behavior. Stanford researchers have created a short, non-humanoid prototype of just such a moving, self-navigating machine.

The robot is nicknamed “Jackrabbot” – after the jackrabbits often seen darting across the Stanford campus – and looks like a ball on wheels. Jackrabbot is equipped with sensors to be able to understand its surroundings and navigate streets and hallways according to normal human etiquette.

The idea behind the work is that by observing how Jackrabbot navigates itself among students around the halls and sidewalks of Stanford’s School of Engineering, and over time learns unwritten conventions of these social behaviors, the researchers will gain critical insight in how to design the next generation of everyday robots such that they operate smoothly alongside humans in crowded open spaces like shopping malls or train stations.

“By learning social conventions, the robot can be part of ecosystems where humans and robots coexist,” said Silvio Savarese, an assistant professor of computer science and director of the Stanford Computational Vision and Geometry Lab.

The researchers will present their system for predicting human trajectories in crowded spaces at the Computer Vision and Pattern Recognition conference in Las Vegas on June 27.

As robotic devices become more common in human environments, it becomes increasingly important that they understand and respect human social norms, Savarese said. How should they behave in crowds? How do they share public resources, like sidewalks or parking spots? When should a robot take its turn? What are the ways people signal each other to coordinate movements and negotiate other spontaneous activities, like forming a line?

These human social conventions aren’t necessarily explicit nor are they written down complete with lane markings and traffic lights, like the traffic rules that govern the behavior of autonomous cars.

So Savarese’s lab is using machine learning techniques to create algorithms that will, in turn, allow the robot to recognize and react appropriately to unwritten rules of pedestrian traffic. The team’s computer scientists have been collecting images and video of people moving around the Stanford campus and transforming those images into coordinates. From those coordinates, they can train an algorithm.

“Our goal in this project is to actually learn those (pedestrian) rules automatically from observations – by seeing how humans behave in these kinds of social spaces,” Savarese said. “The idea is to transfer those rules into robots.”

Jackrabbot already moves automatically and can navigate without human assistance indoors, and the team members are fine-tuning the robot’s self-navigation capabilities outdoors. The next step in their research is the implementation of “social aspects” of pedestrian navigation such as deciding rights of way on the sidewalk. This work, described in their newest conference papers, has been demonstrated in computer simulations.

“We have developed a new algorithm that is able to automatically move the robot with social awareness, and we’re currently integrating that in Jackrabbot,” said Alexandre Alahi, a postdoctoral researcher in the lab.

Even though social robots may someday roam among humans, Savarese said he believes they don’t necessarily need to look like humans. Instead they should be designed to look as lovable and friendly as possible. In demos, the roughly three-foot-tall Jackrabbot roams around campus wearing a Stanford tie and sun-hat, generating hugs and curiosity from passersby.

Today, Jackrabbot is an expensive prototype. But Savarese estimates that in five or six years social robots like this could become as cheap as $500, making it possible for companies to release them to the mass market.

“It’s possible to make these robots affordable for on-campus delivery, or for aiding impaired people to navigate in a public space like a train station or for guiding people to find their way through an airport,” Savarese said.

The conference paper is titled “Social LSTM: Human Trajectory Prediction in Crowded Spaces.” See conference program for details.

Source: Stanford University News Service

ight behaves both as a particle and as a wave. Since the days of Einstein, scientists have been trying to directly observe both of these aspects of light at the same time. Now, scientists at EPFL have succeeded in capturing the first-ever snapshot of this dual behavior.
Credit:EPFL

Entering 2016 with new hope

Syed Faisal ur Rahman


 

Year 2015 left many good and bad memories for many of us. On one hand we saw more wars, terrorist attacks and political confrontations, and on the other hand we saw humanity raising voices for peace, sheltering refugees and joining hands to confront the climate change.

In science, we saw first ever photograph of light as both wave and particle. We also saw some serious development in machine learning, data sciences and artificial intelligence areas with some voices raising caution about the takeover of AI over humanity and issues related to privacy. The big question of energy and climate change remained a key point of  discussion in scientific and political circles. The biggest break through came near the end of the year with Paris deal during COP21.

The deal involving around 200 countries represent a true spirit of humanity to limit global warming below 2C and commitments for striving to keep temperatures at above 1.5C pre-industrial levels. This truly global commitment also served in bringing rival countries to sit together for a common cause to save humanity from self destruction. I hope the spirit will continue in other areas of common interest as well.

This spectacular view from the NASA/ESA Hubble Space Telescope shows the rich galaxy cluster Abell 1689. The huge concentration of mass bends light coming from more distant objects and can increase their total apparent brightness and make them visible. One such object, A1689-zD1, is located in the box — although it is still so faint that it is barely seen in this picture. New observations with ALMA and ESO’s VLT have revealed that this object is a dusty galaxy seen when the Universe was just 700 million years old. Credit: NASA; ESA; L. Bradley (Johns Hopkins University); R. Bouwens (University of California, Santa Cruz); H. Ford (Johns Hopkins University); and G. Illingworth (University of California, Santa Cruz)
This spectacular view from the NASA/ESA Hubble Space Telescope shows the rich galaxy cluster Abell 1689. The huge concentration of mass bends light coming from more distant objects and can increase their total apparent brightness and make them visible. One such object, A1689-zD1, is located in the box — although it is still so faint that it is barely seen in this picture.
New observations with ALMA and ESO’s VLT have revealed that this object is a dusty galaxy seen when the Universe was just 700 million years old.
Credit:
NASA; ESA; L. Bradley (Johns Hopkins University); R. Bouwens (University of California, Santa Cruz); H. Ford (Johns Hopkins University); and G. Illingworth (University of California, Santa Cruz)

Space Sciences also saw some enormous advancements with New Horizon sending photographs from Pluto, SpaceX successfully landed the reusable Falcon 9 rocket back after a successful launch and we also saw the discovery of the largest regular formation in the Universe,by Prof Lajos Balazs, which is a ring of nine galaxies 7 billion light years away and 5 billion light years wide covering a third of our sky.We also learnt this year that Mars once had more water than Earth’s Arctic Ocean. NASA later confirmed the evidence that water flows on the surface of Mars. The announcement led to some interesting insight into the atmospheric studies and history of the red planet.

In the researchers' new system, a returning beam of light is mixed with a locally stored beam, and the correlation of their phase, or period of oscillation, helps remove noise caused by interactions with the environment. Illustration: Jose-Luis Olivares/MIT
In the researchers’ new system, a returning beam of light is mixed with a locally stored beam, and the correlation of their phase, or period of oscillation, helps remove noise caused by interactions with the environment.
Illustration: Jose-Luis Olivares/MIT

We also saw some encouraging advancements in neurosciences where we saw MIT’s researchers  developing a technique allowing direct stimulation of neurons, which could be an effective treatment for a variety of neurological diseases, without the need for implants or external connections. We also saw researchers reactivating neuro-plasticity in older mice, restoring their brains to a younger state and we also saw some good progress in combating Alzheimer’s diseases.

Quantum physics again stayed as a key area of scientific advancements. Quantu

ight behaves both as a particle and as a wave. Since the days of Einstein, scientists have been trying to directly observe both of these aspects of light at the same time. Now, scientists at EPFL have succeeded in capturing the first-ever snapshot of this dual behavior. Credit:EPFL
ight behaves both as a particle and as a wave. Since the days of Einstein, scientists have been trying to directly observe both of these aspects of light at the same time. Now, scientists at EPFL have succeeded in capturing the first-ever snapshot of this dual behavior.
Credit:EPFL

m computing is getting more closer to become a viable alternative to current architecture. The packing of the single-photon detectors on an optical chip is a crucial step toward quantum-computational circuits. Researchers at the Australian National University (ANU)  performed experiment to prove that reality does not exist until it is measured.

There are many other areas where science and technology reached new heights and will hopefully continue to do so in the year 2016. I hope these advancements will not only help us in growing economically but also help us in becoming better human beings and a better society.

 

 

 

 

 

Collecting just the right data : MIT Research

When you can’t collect all the data you need, a new algorithm tells you which to target.

 Larry Hardesty | MIT News Office 


Much artificial-intelligence research addresses the problem of making predictions based on large data sets. An obvious example is the recommendation engines at retail sites like Amazon and Netflix.

But some types of data are harder to collect than online click histories —information about geological formations thousands of feet underground, for instance. And in other applications — such as trying to predict the path of a storm — there may just not be enough time to crunch all the available data.

Dan Levine, an MIT graduate student in aeronautics and astronautics, and his advisor, Jonathan How, the Richard Cockburn Maclaurin Professor of Aeronautics and Astronautics, have developed a new technique that could help with both problems. For a range of common applications in which data is either difficult to collect or too time-consuming to process, the technique can identify the subset of data items that will yield the most reliable predictions. So geologists trying to assess the extent of underground petroleum deposits, or meteorologists trying to forecast the weather, can make do with just a few, targeted measurements, saving time and money.

Levine and How, who presented their work at the Uncertainty in Artificial Intelligence conference this week, consider the special case in which something about the relationships between data items is known in advance. Weather prediction provides an intuitive example: Measurements of temperature, pressure, and wind velocity at one location tend to be good indicators of measurements at adjacent locations, or of measurements at the same location a short time later, but the correlation grows weaker the farther out you move either geographically or chronologically.

Graphic content

Such correlations can be represented by something called a probabilistic graphical model. In this context, a graph is a mathematical abstraction consisting of nodes — typically depicted as circles — and edges — typically depicted as line segments connecting nodes. A network diagram is one example of a graph; a family tree is another. In a probabilistic graphical model, the nodes represent variables, and the edges represent the strength of the correlations between them.

Levine and How developed an algorithm that can efficiently calculate just how much information any node in the graph gives you about any other — what in information theory is called “mutual information.” As Levine explains, one of the obstacles to performing that calculation efficiently is the presence of “loops” in the graph, or nodes that are connected by more than one path.

Calculating mutual information between nodes, Levine says, is kind of like injecting blue dye into one of them and then measuring the concentration of blue at the other. “It’s typically going to fall off as we go further out in the graph,” Levine says. “If there’s a unique path between them, then we can compute it pretty easily, because we know what path the blue dye will take. But if there are loops in the graph, then it’s harder for us to compute how blue other nodes are because there are many different paths.”

So the first step in the researchers’ technique is to calculate “spanning trees” for the graph. A tree is just a graph with no loops: In a family tree, for instance, a loop might mean that someone was both parent and sibling to the same person. A spanning tree is a tree that touches all of a graph’s nodes but dispenses with the edges that create loops.

Betting the spread

Most of the nodes that remain in the graph, however, are “nuisances,” meaning that they don’t contain much useful information about the node of interest. The key to Levine and How’s technique is a way to use those nodes to navigate the graph without letting their short-range influence distort the long-range calculation of mutual information.

That’s possible, Levine explains, because the probabilities represented by the graph are Gaussian, meaning that they follow the bell curve familiar as the model of, for instance, the dispersion of characteristics in a population. A Gaussian distribution is exhaustively characterized by just two measurements: the average value — say, the average height in a population — and the variance — the rate at which the bell spreads out.

“The uncertainty in the problem is really a function of the spread of the distribution,” Levine says. “It doesn’t really depend on where the distribution is centered in space.” As a consequence, it’s often possible to calculate variance across a probabilistic graphical model without relying on the specific values of the nodes. “The usefulness of data can be assessed before the data itself becomes available,” Levine says.

Reprinted with permission of MIT News (http://newsoffice.mit.edu/)