There is a problem with the robots.
Professor Magnus Egerstedt eyes them with dismay. His robots, rotund little things that wouldn’t look out of place in a Roomba ad, are misbehaving. They’re crawling toward two silver cases in the middle of the floor, but just before reaching them, they’re supposed to sense the obstacles and skitter away.
The demonstration is the highlight of Egerstedt’s lecture; it is the center ring of this particular circus. A professor in the School of Electrical and Computer Engineering (ECE), Egerstedt is teaching a class on robot control, and the point of the course tends to be lost if the robots in question are going rogue.
Toward the cases they go, but wait! They’re sliding this way and that way. They come to halting stops or turn dizzily, and some of them don’t recognize the cases at all and slam into their shiny edges. HAL 9000 these robots are not, but their predicament is looking dire.
It’s as if they’re drunk — and in a way, they are. The robots, all Khepera IIIs, have infra-red sensors that are sensitive to light, and they’re suffering beneath the harsh glare of studio lamps. This isn’t a typical classroom lecture — these robots are starring in a massive open online course, or MOOC.
Egerstedt is teaching in a familiar way, but the setting is different. There are no students present, and instead of a classroom or lecture hall, the lesson is being given in a studio with a jungle of lamps overhead. Every move Egerstedt makes is recorded for online broadcast, and his robots are not taking kindly to the extra light. Frustrated, Egerstedt calls out to Brian Wilson, the instructional media producer who is filming the lesson.
“Brian, can we do it again?” he asks.
“I’m just going to keep rolling,” Wilson replies.
“Yeah,” Egerstedt says. “Till it works.”
And it does work eventually, and when it does, the effect is sublime. Egerstedt turns triumphantly to his student assistant, Jean-Pierre de la Croix.
“Yeah!” he exclaims. “Robotics is so easy, man.”
A MOOC is a special type of online class. Available to anyone with an Internet connection, most MOOCs offer college-level material via pre-taped lectures, and course enrollments can swell to tens of thousands of students. The premise is simple: Bring higher education to the masses by cutting through academia’s red tape and, most tantalizingly, offering many courses (including Egerstedt’s) at the low, low cost of free.
Like other universities, Georgia Tech has embraced the promise of MOOCs, and the Institute offers them across a growing array of disciplines. But a newly hatched breed of pedagogy brings risk, and as MOOCs multiply around the country, observers express both excitement and skepticism toward them.
Egerstedt is a believer, and he’s also a guinea pig. Few engineering MOOCs have sprouted thus far. There is no blueprint for this course, Control of Mobile Robots, and no backup plan when Egerstedt’s recording sessions, like the one with the silver cases, do not go quite as expected.
None of that dampens his enthusiasm. He wants to be, he says, a Mick Jagger-like figure of the robotics world.
Can he pull it off? Can he become a rock star of online education, spreading the gospel of robotics to students around the globe? More pressing is the question of what students will take from his lectures: No one who enrolls in his course, even those who do exceptionally well, will receive university credit for their efforts.
Still in their infancy, MOOCs are stirring debate in newspapers, magazines and blogs galore. This is the story of one MOOC — and the man who would be Mick Jagger.
“A chance to do good”
There is perhaps no one better suited to teach a MOOC than Magnus Egerstedt.
He is impossible to dislike. His excitement for his work is palpable. And, maybe most important, he is unafraid of the unexpected.
“I am not an expert on MOOCs,” he said. “I just happen to be doing it.”
Egerstedt, originally from Sweden, has worked at Georgia Tech for 12 years. Recently awarded ECE’s first-ever Schlumberger professorship, he typically teaches senior- and graduate-level courses on robotics and controls, and robots are the epicenter of almost everything he studies. His CV features work with self-driving cars and humanoid robots, but his recent research focuses on swarm robotics.
Ask him to tell you about it sometime; when he discusses his latest projects, he's animated and eager to explain details. Through his work in swarm robotics (or, to put it a simpler way, robots moving in groups), Egerstedt studies animal behavior, applying lessons from the natural world to the artificial one.
"I've always been fascinated by schooling fish and bird flocks," he said.
There is strength in numbers. The power of a single robot — even an expensive one — can be dwarfed by dozens of cheaper beings' capabilities. Imagine, for example, the tattered landscape left behind by a big earthquake. In the rush to find victims, a hundred inexpensive robots might be able to comb through wreckage faster and more efficiently.
“The question there is cooperation,” Egerstedt explained. This is the sort of thing he could talk about all day. He is fueled first and foremost by Starbucks coffee, but he also thrives on an unsinkable enthusiasm for his work.
Which, in a way, explains his foray into MOOCs. Egerstedt loves teaching robotics, and he decided MOOCs could give him the chance to share that passion with thousands.
“There is something fundamentally satisfying,” he said, “about having all these people across the world learning something that they otherwise wouldn’t learn.”
Egerstedt was first yanked into the MOOC world in early 2012, when he heard a talk by Udacity cofounder Sebastian Thrun. Udacity is one of the three major startups offering MOOCs; the others are edX and Coursera (the platform that eventually hosted Egerstedt’s MOOC). Michael McCracken, director of online course development and innovation for the Institute’s Center for 21st Century Universities (C21U), learned that Egerstedt had been contemplating a MOOC of his own.
By this point, Tech had a contract with Coursera, and Egerstedt had laid the foundations for the class he hoped to teach.
“I knew the story I wanted to tell,” he said. And so his course was born.
He began filming lectures and demonstrations on robot control in Fall 2012. Collaborating with one of Georgia Tech Professional Education’s instructional designers, as well as C21U, Egerstedt created a seven-week course broken into bite-size pieces. Each week of material was given its own theme and divided into eight or so “sublectures” between about five and 11 minutes long.
In addition to their cosmetic distinctions, MOOCs differ from traditional classes in their basic structure. If you take an on-campus course at Georgia Tech, you could find yourself in a lecture hall for hours at a time, listening to a professor cover multiple topics during a single session. Tech MOOCs, meanwhile, demand content in a more easily digestible form. Forty-five minutes to present a new concept? Nope — you’ve got just 10. To ensure students are grasping the material, each week’s lessons are capped with a quiz.
One of the most vocal proponents of this teaching model is Rich DeMillo, the director for C21U (and one of Georgia Tech’s key MOOC experts). In 2011, he published a book called “Abelard to Apple: The Fate of American Colleges and Universities,” which examines the future of higher education. It has a chapter about MOOCs, which DeMillo originally thought was “a very way-out concept.”
After the book’s release, he continued investigating their usefulness and efficiency, and one of the things he appreciates most about GT MOOCs is the way they are delivered. Teaching lessons in small doses, he says, incorporates the principles of “mastery learning,” a concept developed by educational theorist Benjamin Bloom. The idea is that everyone can learn when content is fed a little bit at a time and students are constantly tested.
“That is better than the normal classroom,” he said, “and we know this because there’s 100 years of educational research that show it.”
Egerstedt admits that after years of educating via longer lectures, he initially struggled to adjust to a new format. But by early 2013, when he was wrapping up MOOC filming, he had fallen into an easy rhythm. Though his PowerPoint slides offered a compass for each lesson, Egerstedt improvised almost all his lectures as he gave them.
The class was due to start Jan. 28. To promote it, Coursera released a trailer of sorts on its website, featuring footage of Egerstedt and some friends: a Khepera, a humanoid robot, and a drone. (Certainly no other MOOC previews boasted the same James Bond flair.)
“I look very much forward to seeing you enrolled in Control of Mobile Robots in the not-too-distant future,” Egerstedt told viewers.
Still, concerns bubbled in his mind. MOOCs are a gamble — some have failed, resulting in suspension or cancellation. And he wondered whether engineering coursework, in which practice can be just as important as theory, might prove unadaptable for MOOCs. Engineers, he noted, “develop theory not because we’re mathematicians, but because theory helps us solve real-world problems. That can get lost in the video shuffle.”
But when he peeled the worries away, Egerstedt found a dozen shimmering opportunities. He wanted enrollees to find intellectual stimulation and think more analytically, but his greatest wish was to catapult students into new worlds, igniting lifelong interests in both robotics and engineering.
“I truly believe there is a chance to do good,” he said.
By the time his MOOC began in January, nearly 40,000 students had registered.
The Making of a MOOC
In 2012, Focus Features released a new film adaptation of Leo Tolstoy’s “Anna Karenina.” The movie hinged on a lavish conceit: Much of the action was filmed as if taking place on a stage.
Grand ballrooms served as sets; elegant décor was reduced to mere props. The overarching plot, however, remained largely the same.
This retooled “Anna” drew mixed reviews. Critics praised its stars’ performances, but some admitted they were turned off by the theatrical motif. In a review for Australia’s Special Broadcasting Service, the critic Don Groves wrote that the film’s “artifice creates an invisible wall between the cast and the viewer, frequently working against the sense of realism and blunting the emotional intensity.”
In a sense, much of the criticism levied against MOOCs is similar. Without a conventional classroom environment, some say, the courses could lack intellectual spark.
Of course, many of the students in Egerstedt’s MOOC might never get to physically sit in his classroom anyway. In the months before the class started, he faced an email deluge from prospective students around the world. Excited messages poured in from South America, from Europe, from Southeast Asia, and the sentiments from across linguistic and geographic borders were the same: We want to take your class!
Other emails were written closer to home. Egerstedt read notes from American undergraduates and even some from younger teenagers, and a few high school robotics groups asked if their members might benefit from the lessons.
Egerstedt found these questions slightly worrisome. It is a truth universally acknowledged that robots are cool, but he had bigger things in mind for his MOOC.
“There is going to be a level of mathematical sophistication,” he said, “that is hard for a student.”
Watering down his material was out of the question. The Coursera website offered a warning of sorts, recommending that potential Control of Mobile Robots students bring a background in calculus and linear algebra. Far from an occasional guest, math played a starring role in lectures and quizzes. Here’s a question pulled directly from one of the MOOC assessments:
In Week 6, we used a point-robot model, ẋ=u, x∈R2, to describe the robot dynamics. But, the actual robot is nonlinear. So, when implementing the behaviors, we have to map the point-robot input u=[u1,u2], onto the real control signals, i.e., onto (v,ω) in the unicycle case. Which of the following control designs would make a unicycle model mimic the point-robot model reasonably well?
v=∥u∥, ω=K(atan(u1/u2)−ϕ), K<0
v=∥u∥, ω=K(atan(u2/u1)−ϕ), K>0
v=∥u∥, ω=K(atan(u1/u2)−ϕ), K>0
v=∥u∥, ω=K(atan(u2/u1)−ϕ), K<0
The course’s point, Egerstedt stressed, was not to simply “build a Lego robot and do some random stuff.” To offer his audience the same level of content that Georgia Tech students would get, challenging math had to remain a staple. But that invites another question: Why should an institution like Georgia Tech (or any university, for that matter) give away its best stuff for free?
As a prominent MOOC advocate, DeMillo has faced that question again and again.
“To me, that is actually the least interesting discussion to have,” he said. The people throwing themselves into MOOCs — the course planners, the professors, the administrators — are “taking the long view.”
“The worst thing you can do is make a short-term opportunistic decision when you’re not quite sure what the 10-year picture is going to look like,” he said.
DeMillo sees Tech translating a portion of its course catalog into MOOC form. The possibilities are enticing: MOOCs could be distributed to high school students, he said, to acclimate them to college work in the same way an Advanced Placement curriculum does. The advantage of using a MOOC over an AP course, however, is that “we can tailor the MOOC to better match the Georgia Tech curriculum.”
On a more philosophical level, DeMillo stresses, the courses can redefine the very idea of what good teaching is. And that’s where Egerstedt comes in.
“Magnus is a very infectious guy,” DeMillo said. “He comes to the subject with a lot of enthusiasm, and it shows up well in these online courses.”
It shows up well in person, too.
Control of Mobile Robots generated excitement around the world, but Georgia Tech students — the ones who are right here in Atlanta and working toward an engineering degree — stood to benefit as well. Egerstedt teaches a course called Embedded and Hybrid Control Systems, or ECE 4555. During past semesters, the class has proceeded as many engineering classes do: Egerstedt devoted some sessions to lecturing and others to hands-on experiments. But for his Spring 2013 group, he decided to shake things up a bit.
Big swaths of ECE 4555 material overlap with Egerstedt’s MOOC. Given that, then, he decided to try a “flipped classroom,” in which students watch lectures online to free up class time for discussion and hands-on work.
Ryan Bahr is one of those students. A bright and chatty Tech senior, Bahr has an easy smile and a penchant for robotics. (You’re apt to find him browsing diydrones.com, a forum for enthusiasts of unmanned aerial vehicles.)
Bahr, an electrical engineering major from Peachtree City, hopes to one day work in radio frequency engineering. ECE 4555 is the second course he has taken with Egerstedt, and when he signed up, he didn’t know online lectures would be part of the experience. Unlike the MOOC, this section of ECE 4555 came with a prerequisite class and built on earlier material.
With Egerstedt’s online lectures teaching the basics, his ECE students were free to spend class time applying their knowledge in robotics experiments. Bahr pointed out the benefits of learning engineering this way.
“In some classes, you just always talk about it,” Bahr said. “Being able to use it once on a real live object makes it 10 times more useful in the real world, in my opinion.”
A s Egerstedt’s MOOC and flipped course progressed, the physical world and the digital one sometimes mirrored each other in unexpected ways.
On a bright Thursday afternoon, a classroom in Tech Square filled with students. A whiteboard, so big it ate up an entire wall, beckoned with equations in red, blue and black marker. There were no desks; the thirty or so students who eventually wandered in bunched together on the floor.
These were the students of ECE 4555, a tiny sampling of Control of Mobile Robots’ thousands of enrollees but the only ones participating in the flipped course. After watching Egerstedt’s lectures online, the students tackled a new robot-control project each week with both Egerstedt and graduate assistants on hand to offer guidance. Tuesdays were for writing code; Thursdays were for testing it on unsuspecting robots.
This week’s assignment: Program a Khepera to travel a predetermined path, hitting a series of goal points marked on the floor with tape. Without desks or chairs, some students sprawled belly-down on the floor like swimming frogs; others leaned against the wall and stared into laptop screens.
Egerstedt, dressed in his typical blazer-and-dark-jeans uniform, opened the class by soliciting questions. Then the experiments began, with graduate assistants Jean-Pierre de la Croix and Amy LaViers standing nearby to help out. (LaViers, another MOOC co-star, led a video lecture each week to help prepare students for quizzes.) Egerstedt worked his way through the room, peering at students’ code and watching their robots jolt to life.
Divided into groups, the students chatted among themselves as they awaited Egerstedt’s evaluation or sent Kheperas zipping around the floor. Between the robots and the budding programmers, the room hummed with energy, which is just the way Egerstedt likes it.
“The turns are way faster than the linear velocity,” he advised one group upon scanning their work, “which means you are going to end up slipping a little bit.” Back to the drawing board: The students huddled to tweak their code.
Another group mastered the assignment on the first try. Their Khepera strutted confidently along its assigned path, while Egerstedt, overjoyed, whipped out his iPhone to shoot video.
“That is picture-perfect!” he gushed. “That was fantastic!”
A big benefit of the flipped class was the immediate feedback it offered Egerstedt. Through his on-campus students, he could quickly gauge how his online lectures were being received — which concepts were well elucidated and which might need some clearing up. But his favorite part of the class, he said, was that it allowed more time for hands-on work with students.
“People seem so engaged,” he said, “and it’s so exciting and the projects are going so well.”
Bahr was enjoying the flipped format, too. He was glad for more time to work with robots, which gave him and his classmates a chance to create their own robotics projects toward the end of the semester.
“Overall it’s really awesome,” he said.
How could the MOOC compete with that? Egerstedt’s video lessons, though laced with his humor, were typically subdued affairs. But the class, he quickly discovered, was much larger than his presentations.
In a lecture for his MOOC’s second unit, Egerstedt appeared against a black backdrop. Seated behind a desk, he addressed his thousands of students about the necessity of testing controls with a simulator.
Toward the end of the lesson, he again used a Khepera that, just like in the ECE 4555 session, was programmed to slide across the floor and hit a goal point. As it moved, Egerstedt offered commentary like a sports analyst.
“As you can see, the turn was nice — there were very few, or little, oscillations here,” he observed.
Cheep, cheep, cheep. The machine ambled toward the tape.
“And the Khepera is making it very nicely all the way to the turquoise goal point,” Egerstedt concluded. “So, we will call that a success.”
Without any students to witness the success in person, his proclamation was greeted with silence. But on the Internet, MOOC participants were buzzing.
On the Control of Mobile Robots website, students took to discussion boards to trade tips, share questions, and chat about the course. Egerstedt, LaViers and de la Croix monitored the forums regularly, looking for chances to offer help or clear confusion. But to Egerstedt’s surprise, students were finding ways to support each other.
If a poster expressed frustration with a concept, another user might chime in to explain it. Just like on other Internet boards, students could “vote up” pertinent questions and “vote down” less relevant remarks. This system helped Egerstedt and his assistants better assess where their expertise was needed.
Meanwhile, in-person study groups sprouted in countries around the world, and on Twitter, students used the hashtag #conrob to discuss the class. A student-run website with extra robotics resources—including YouTube videos, math help and a Khepera simulator—appeared as well.
At Georgia Tech, Egerstedt directs a lab called the Georgia Robotics and Intelligent Systems Laboratory, or GRITS Lab. The name of the student-operated site: GRITS Group.
“It’s quite wonderful,” Egerstedt said, “to see this community emerging.”
The Golden Opportunity
Five days a week, Collin Brown drives to work at a small industrial park in suburban Cincinnati. In his job as a software engineer, he sifts through computer code and debugs programs.
But for two months, Brown devoted his precious weekend hours to another type of engineering. As a sort of extracurricular pursuit, he enrolled in the Control of Mobile Robots class and found himself enamored with the material.
“The class really helped bridge theory to reality and then to practice,” he said. “Now I want to learn more and run out and build my own robot.”
A graduate of Tech’s Guggenheim School of Aerospace Engineering, Brown is also a science lover, a trivia buff and curious about the mechanisms behind everything. Participating in Egerstedt’s MOOC, he said, offered a way to learn something new and build his resume at the same time.
Even though the class didn’t award official credit, students who completed it successfully (averaging a score of 60 percent or above on all quizzes) received a certificate from Coursera. Brown, who liked to gobble several videos at a time and aced nearly every quiz, earned a seal of distinction on his certificate to honor his performance.
And he wasn’t the only one.
By the time Control of Mobile Robots ended in mid-March, nearly 4,300 people had passed it, and about half of those did well enough to merit the seal of distinction. Participation levels fluctuated during the course’s seven weeks, but when the MOOC closed, more than 40,000 people were registered. (Not everyone who enrolled tried the quizzes — some just wanted to watch lecture videos, read forums or observe unnoticed.)
Egerstedt noted that teaching a MOOC allowed him to reach more students in seven weeks than he had in his entire Georgia Tech career.
“This is fun, and you have impact beyond your wildest dreams,” he said.
Teaching advanced engineering to a class of thousands is bound to attract some attention. Even before the MOOC ended, its successes were drawing notice: With a higher-than-usual retention rate and plenty of positive feedback in online forums, Control of Mobile Robots became a beacon of the MOOC universe.
Egerstedt was invited to speak at conferences around the country, which he found “satisfying and a little surreal.” In March, he was one of three featured speakers at a conference in San Diego. The other two lecturers: Coursera’s founder and the director of edX.
Though his class was quickly trumpeted as a model for engineering MOOCs, Egerstedt tried to maintain a realistic tone in his speeches. He didn’t hide, for example, the fact that a few participants fired personal attacks at him and his teaching assistants early in the course. He didn’t mind being a target himself, but seeing his graduate students subjected to inflammatory remarks was upsetting.
“That pissed me off to no end,” he said.
But the negativity melted fast, and by the course’s end, its forums hosted praise for both Egerstedt and his graduate students. Though he earned a kind of personal fame, Egerstedt wanted to place his victories in a broader context: He hoped his class’ success would prove that not only were MOOCs and flipped courses viable, but that they could work with concepts steeped in high-level math and engineering.
“There is appetite for serious content,” Egerstedt said. “There is a clear niche for these upper-level engineering courses.”
That’s where he thinks Georgia Tech and the College of Engineering can shine.
Universities around the world are scrambling to construct their own MOOCs, and wading through the increasingly crowded pool can be daunting. But Tech specializes in engineering classes that teach highly sought skills. With the right finesse, Egerstedt maintains, GT can create MOOCs that few other institutions can match, and that in turn will strengthen the Institute’s brand around the globe.
“It’s a golden opportunity we should take,” he said.
His Control of Mobile Robots students seemed to agree. As the course drew to an end, a discussion thread about a follow-up class materialized, and it grew into one of the board’s most popular threads. Its title: “WE WANT CONTROL OF MOBILE ROBOTS 2!!!!”
Dozens of participants signed their names to pleas for a second class, with some listing their home cities as well. The geographic diversity was astonishing: Students came from Savona, Italy; Accra, Ghana; Montreal, Quebec; Lima, Peru; Karachi, Pakistan; Guanajuato, Mexico; Boulder, Colorado. (And the list went on.)
Egerstedt is indeed planning a second course, and he’s also making a few revisions to the first one so that it can be offered again. (The University of Hawaii has expressed interest in using it for its own flipped class.) Not too long from now, thousands of new enrollees will be learning about robot control from Egerstedt.
Some of his inaugural MOOC students, meanwhile, were reluctant to say goodbye. In the final days of the course, they flooded the forums with thank-you notes for Egerstedt, de la Croix and LaViers. A few of the messages posted:
- “I really loved this course! Excellent material and really fun to follow. Thank you Professor Egerstedt and all other people involved in this project too.”
- “Your obvious enjoyment of the topic was inspiring, even awesome.”
- “Thank you Prof Magnus Egerstedt, Amy and JP. It was great and fun learning from you guys.”
- “It's been 40 years since I last looked at modern control theory. If I'd had a teacher like the Prof all those years ago, who knows how my life would have changed.”
And then, tellingly, one student wrote this:
“Dr. ME = Rock Star!!! Keep changing the world for the better by equalizing education...”