Author: dsilver829

I love self-driving cars!

Win an Interview with Lyft’s Self-Driving Car Team

For the last few weeks Udacity and Lyft’s Level 5 Engineering Center have been hosting the Lyft Perception Challenge to identify the best Udacity students for interviews on Lyft’s self-driving car team.

The competition involves segmenting driving images, like the image above. If you’re in or have graduated from the Self-Driving Car Nanodegree Program and interested in working at Lyft, you should participate! And if you’re not in the Nanodegree Program, you should join 🙂

The competition closes on June 1, 2018, so there are still a few days left to enter your submission.

If you’re looking for an approach to get started, my Udacity colleague Kyle Stewart-Frantz has posted a terrific walkthrough of his deep learning approach to the challenge.

Kyle trained a fully-convolutional network using data from the CARLA simulator, and then found that the data applied pretty well to real world camera data.

https://youtu.be/K8W_MzmooI0

“Deep Reinforcement Learning Doesn’t Work Yet”

The blog post, “Deep Reinforcement Learning Doesn’t Work Yet”, has been making the rounds for the last few months, but I only just sat down to read it.

It’s a terrific summary of the current state of deep learning research, reasons why DeepRL is not yet living up to its hype, and hope for the future. It’s written by Alexander Irpan, who works on RL at Google Brain.

“For purely getting good performance, deep RL’s track record isn’t that great, because it consistently gets beaten by other methods.”

There is a lot of interest in using DeepRL for self-driving cars. While this is a super-exciting opportunity in theory, in practice DeepRL has not been effective.

“I tried to think of real-world, productionized uses of deep RL…The way I see it, either deep RL is still a research topic that isn’t robust enough for widespread use, or it’s usable and the people who’ve gotten it to work aren’t publicizing it. I think the former is more likely.”

A big challenge for RL generally, and particularly when it comes to self-driving cars, is the design of a reward function. It’s not clear what the reward function for driving a car would be. And, as Irpan makes clear, unless the reward function is designed near perfectly, the learning agent is going to find all sorts of disastrous shortcuts to maximize the learning function at the expense of violating the implicit rules of the game.

“A friend is training a simulated robot arm to reach towards a point above a table. It turns out the point was defined with respect to the table, and the table wasn’t anchored to anything. The policy learned to slam the table really hard, making the table fall over, which moved the target point too. The target point just so happened to fall next to the end of the arm.”

Irpan is hopeful about the future of RL for practical problems, but cautiously so. Definitely worth a read.

A New Country for Self-Driving Cars

Intel-owned Mobileye has announced they have a fleet of 100 self-driving vehicles on the roads in Jerusalem. That’s a lot of cars.

Israel has been a hotbed of autonomous vehicle development, but Israeli automotive companies face the same problem that almost all Israeli companies must surmount — a small local market. Mobileye has grown into a world leader in computer vision by supplying its perception systems to large automotive manufacturers outside of Israel.

Nonetheless, Mobileye founder and CEO Amnon Shashua writes that testing in Israel is important because of the challenges of Israeli driving. The title of the post is, “If You Can Drive in Jerusalem You Can Drive (Almost) Anywhere”.

“Jerusalem is notorious for aggressive driving. There aren’t perfectly marked roads. And there are complicated merges. People don’t always use crosswalks. You can’t have an autonomous car traveling at an overly cautious speed, congesting traffic or potentially causing an accident. You must drive assertively and make quick decisions like a local driver.”

Mobileye also published a video showing its self-driving cars executing several lane changes and other maneuvers in heavy traffic. The accompanying description of how the Mobileye vehicle is signaling to other cars on the road is impressive.

I was particularly struck by the description of Mobileye’s planning function.

“The part of our driving policy system that proposes actions is trained offline to optimize an assertive, smooth and human-like driving style. This is a proprietary software developed using artificial intelligence-based reinforcement learning [DS: my emphasis] techniques. This system is the largest advancement demonstrated in the fleet, and you can see the impressive results in the event visuals.”

Reinforcement learning has led to tremendous breakthroughs in games like chess and Go. It has stalled in its ability to translate to the real world, where simulation is less high-fidelity. This marks one of the first successful deployments of reinforcement learning to physical world problems.

Lots of neat stuff here.

How Do You Stop a Self-Driving Car?

When in doubt, use scissors.

* Open one door to prevent the self-driving car from moving
* Break a window if doors are locked and immediate entry to the vehicle is necessary
* Call Waymo to unlock the doors remotely if there is time
Keep at least one door open until the vehicle engine is turned off (Step 7) or 12V Cut Loop under the hood is cut (Step 8)

There are several more options that require scissors, with pictures, all for emergency responders.

Rural Self-Driving Car from MIT

A research team from MIT built a self-driving car to operate on rural roads, without the need for high-definition maps. HD maps are one of the big limiting factors for self-driving cars, which need the maps in order to localize themselves on the road with near-exact precision.

Unfortunately, HD maps are super-expensive to build and maintain, so any progress on this front is a big deal.

The MIT project, MapLite, uses a combination of open-street map data and lidar data to navigate through unmapped environments. Instead of trying to build a map as it goes (a technique known as “simultaneous localization and mapping”), the vehicle merely uses lidar to identify flat, drivable surfaces, and then identifies a viable trajectory to reach its next waypoint.

All of this seems pretty straightforward, but what really struck me was a seemingly obvious thing they mention in the PR writeup.

“I imagine that the self-driving cars of the future will always make some use of 3-D maps in urban areas,” says [MIT graduate student Teddy] Ort. “But when called upon to take a trip off the beaten path, these vehicles will need to be as good as humans at driving on unfamiliar roads they have never seen before. We hope our work is a step in that direction.”

I had always kind of assumed that vehicles relying on HD maps would be a stuck within their geofences, but of course that’s not necessarily the case. A vehicle could use HD maps in urban areas, and then a non-HD map system in other areas. And of course there could be segments of the world in which neither approach is viable and thus those areas would be completely off-limits (until the mapping gets done, say).

Totally obvious when I write it down, but not something I had ever quite pieced together on my own.

Bangalore Traffic and Self-Driving Cars

I’ve been in Bangalore, India, for the last two days and it has been a delightful trip. The traffic here is fiercesome, though. Even within India, Bangalore has a reputation for congestion on the roads.

Many of the Indians I’ve spoken with look at the traffic here and express skepticism that self-driving cars will ever come to India, or at least any time soon.

I suspect self-driving cars will arrive in India sooner than people expect, though. The key will be that the first experiments will start just as they did in the United States: in locations chosen specifically for their ease of driving.

For years, self-driving cars in the US only operated in Silicon Valley and Las Vegas. And there was a reason for that. Those locations are flat and sunny, and the streets are rectilinear. Even now, Waymo’s first public rollout is in Phoenix, which is flatter, sunnier, and more rectilinear still.

Something similar will happen in India. Self-driving cars won’t come to Bangalore first — that would be crazy. They’ll come first to private campuses and controlled-access public roads with good pavement and markings, where the challenge of getting self-driving cars to work will be more manageable.

Over time, self-driving cars will expand their footprint, in India and elsewhere, but there will be learning curve. Just as self-driving cars in the US have now reached Detroit and Pittsburgh and even Boston, eventually self-driving cars will get to Bangalore.

But you have to crawl before you can walk 😉

Meet Me in Bengaluru (and Delhi!)

Virat Kohli is meeting us in Bangalore, too. (Just kidding.)

I am excited to travel to India for the first time tomorrow! I’ll be visiting Bangalore, Mysore, Chennai, and Delhi.

My colleagues at Udacity India are helping us to host two student events while I am in India.

On Sunday, May 6, we will be holding a Self-Driving Car Networking Event at 91springboard in Bengaluru from 11am to 1pm. Please come to say hello!

And on Friday, May 11, there will be another Self-Driving Car Networking at the Udacity India office in Delhi.

If you are a Udacity student or graduate, I hope to see you at one (or both, if you are ambitious!). And I am excited to experience India!

Update: I had the day of the week wrong for our event in Bengaluru. That will be on Sunday, not Monday.

Lyft Hiring Challenge

Lyft is serious about self-driving cars. Last fall, they announced they were funding 400 scholarships for Udacity’s Intro to Self-Driving Cars Nanodegree Program, with a focus on increasing diversity in the autonomous vehicle industry.

Today, Lyft and Udacity announced the Lyft Perception Challenge to identify top Udacity students to interview for positions on Lyft’s Level 5 self-driving car team.

The project requires students to use computer vision techniques to identify and locate vehicles on the road. These are the same skills taught in the Udacity Self-Driving Car Engineer Nanodegree Program. We are excited to see what students develop even above and beyond what we teach in class!

Udacity Self-Driving Car students and alumni are invited to participate. The top 25 students with US work authorization will earn interviews with Lyft’s team. The top 150 candidates from around the world will be invited to special interview preparation workshops with Udacity’s Career Services team.

Our goal on the Udacity Self-Driving Car Team is to help connect as many students to jobs as possible. We are excited to be working with Lyft to achieve that goal and make self-driving cars a reality for everyone.