Big data digest works  

author: Long Mu Xue Wei Zi min

If we want to build an unmanned fleet, Sure7x24 Hours on the road, What preparations are needed? Under the current technology, Can unmanned vehicles accomplish this challenge? There's just been an attempt, Name it“ Driverless marathon project”(The
Drive-a-thon), And recorded the thrill in detail“24 Hourly driving diary”.

Last month, Silicon Valley's mysterious self driving start-up A driverless car24 Hour Marathon, In order to understand how to, Maximize the driving time of unmanned vehicles.

before, Many self driving companies have tried to make unmanned vehicles try out in different periods of the day and different road conditions.
We hope to explore further, At maximum consumption, How to operate a self driving“ Fleet”. Founded in Silicon Valley two years ago, Take deep learning as the starting point to promote the automatic driving technology, At present, its technology has been iterated to the fourth generation, ReachedL4 level, That is to say, highly automatic driving. Last year6 Announce completion
5000 Ten thousand dollars B Round financing.

Wu Enda, former chief scientist who left Baidu last year (Andrew Ng) Also Board of directors. His wife Carol Reiley
Is one of the founders of the company, Company incumbent CEO Sameep Tandon His students at Stanford. Take this.24 Hour“ Physical exertion” Project is called“ Driverless Marathon”, And in overseas mediamedium An article was published on“24 Hours driverless driving diary”, It records the experience of the whole project from preparation to driving.
Big data abstracts are organized as follows.( Original link:

Driverless Marathon!

First, We set ambitious goals:

Autopilot at least400 Mile;

Overcome all kinds of road conditions and weather;

Ensure that90% Autonomous normal operation in time( We define autonomous normal operation as having drivers, Automatic system on, door closed, Vehicles on the road);

Zero accident, Zero accident, Zero accident. Say the most important thing three times!

Our speed limits on urban and suburban streets are25 reach40 Mile, therefore400 Miles is not an easy goal: We can't101-S Set cruise mode on Highway, Then let the car run.

The entire event takes place on the ground streets of mountain view, California: from11 month16 Morning9 From time to time2017 year11 month17 Morning9 Time.

To ensure the smooth running of the marathon, Many of our employees spent the night in the Office.

We have one24 Autonomous vehicle driving continuously for hours“ Main force“, There are other vehicles, Formed a team. Autonomous driving is a key part of our business model, It's also a key part of our technology stack, So we managed to pick up staff, Friends and family areDrive
ai Another goal of Marathon.

We recently launched an upgraded version of the driving app,Drive-a-thon Provides a good opportunity for internal testing of new use cases. Limits of the autopilot program

Today, Even if it's a driverless vehicle that works continuously24 hour, It also requires a lot of careful planning. Not to mention overcoming all kinds of weather and road conditions.

First, Whenever? On the California Highway, We all have to have a trained and safe driver in the car according to the law. Our safe drivers need to have higher attention and driving power than ordinary drivers on the actual road.

however, A driver can't24 Keep alert for hours. We have assigned five well-trained safety drivers to escort the team.

Except for the driver, Our mechanical staff is still Headquarters waiting, Be ready to deal with anything from a flat tire to a faulty sensor. The infrastructure team that we just upgraded the data logging system is on standby.

Data channel is an underrated problem in unmanned vehicle operation: Through a variety of sensors and remote sensing technology, Our drone records gigabytes of data per minute.
For typical road test, We have plenty of car storage. And for endurance, This means“ Hot switching”, The recording memory can be replaced without shutting down the system.

Last, We hope this activity will be interesting. We invited the whole company, There are many friends and family to experience this activity.

24 Journal of the hour Marathon

Murphy's law came true the morning of the event: It rained cats and dogs when we woke up, It's expected to last all day. Although from the beginning of this year, We did a lot of rain testing in the rain, But it's the most thorough shower we've ever had. For all that, We did it, too.

morning9 spot, Here we go .

New safe driver, The same rain▲

One day, We have carried out dozens of times through the internal sharing ride application, And keep our90% Goal of autonomous uptime.

In the evening, The torrential rain has lightened, But it's getting darker around us. From the perspective of perception, Driving at night is more challenging.
But the better it is, the later it gets to night, The less traffic we have on the road, Can cover more distance every hour.

Tent in our office▲

After midnight, We decided to go back to the Office. We set up an indoor camp, Accommodation for all our staff, Order pizza, Play a game, And cheer for the operation team working late at night.

The next morning, The road is calm, But we still have challenges.
It is very important to keep the driver's safety alert, therefore, We've got another passenger on board all the way, Photograph, Play music, And of course enjoy our car system!

Night Cruise▲

As the sun rises, There are more cars on the road, We got to the early peak.

From seven to nine in the morning, On the 24th hour of autopilot, The car is still running smoothly, It's amazing. For any vehicle, travel24 hour( It's only for natural gas in the middle5 Minute), It's all amazing. Despite the challenges, We still did:
Hold sensor, Computer, Display and auxiliary system in the whole24 Hours of activity. morning9 Click a little. We went back to the garage at headquarters. Successful Marathon!

Result data

Let's review our goals and objectives:90% Time independent normal operation,400 Automatic mileage, Zero accident.

stay24 Within hours, Our car is in autonomous mode22 hour40 Minute, stay94% In the time of. The main reason we stopped on the way was to change the safety driver( about35 Minute), Replace data drive and related software maintenance( about25 Minute), Resolve a software error( about15 Minute) And supplement natural gas( about5 Minute). When we are in11 month17 Morning9 When the point passes through the end point, We've already opened410 Mile, That means our average speed( Parking time on the way is also included) yes17 Mile/ hour. Last, most important of all, We finished the activity safely.

viewpoint: How to operate an unmanned fleet

The development of driverless vehicles not only changes our mode of transportation, And change the world: In the coming decades, Land use, Infrastructure, Insurance industry, There will be fundamental changes in wireless communication and other fields. The industry of driverless vehicles is not just about making driverless vehicles, And it's about a driverless ecosystem.

Let an unmanned fleet run efficiently, There are many factors to consider:

Change parking lot into maintenance center:
The average daily parking time of ordinary vehicles is22 More than hour, The goal of the unmanned fleet is80% All the time on the road. Although this means that the percentage of vehicles parked at any given time is relatively low, But there are different parking requirements for driverless vehicles. The unmanned fleet will need a dedicated maintenance center as a charging or gas station, data warehouse, Sensor calibration station, Cleaning facilities, etc. Prepare for these big changes!

Pick up passengers:
Self driving is known as the solution“ Last kilometer” A panacea for problems, But there are many aspects of the taxi experience that we take for granted. How do you make sure the right person has entered the vehicle? If one or more people need help carrying luggage, What should I do?? What if a person has a disability? this“ The last meter” The problem of the unmanned vehicle fleet is an important factor to consider. Finding a safe place to park is often overlooked by driverless vehicles. Is it acceptable to block the lane or bicycle lane for a few minutes? If there are docking points, We need to detect the edge of the road correctly, Ideally, do not dock in the red paint area. But if there is no legal docking point, What should we do? It is a common practice to discuss and develop a contingency plan with the driver( Buddha trip: Master, don't move, I came over.), But robots“ Driver” It's still strange.

AV + EV(Autonomous + Electric vehicle):
Electrification will undoubtedly play an important role in the future of autonomous driving technology. The energy cost of the hybrid platform we use in driving is about0.09 dollar/ Mile, Like TeslaModel
S Today's electric cars are closer to a mile0.04 dollar. however, Before charging and battery replacement technology is further mature, Hybrid is the best choice. Drones want to reach90% Utilization, It's impossible to charge all night. Electric vehicles have not been fully adopted, But we think, The introduction of electric vehicles will become an important force in improving the infrastructure needed for the next generation of energy storage technology. The real demand for unmanned motorcades comes from enterprises, This is also Already positioned atToB Another reason for.

Today, In many areas, unmanned vehicles are required to be equipped with trained and safe drivers, Whether it's in the car, Or remote monitoring of unmanned vehicles. New jobs will follow, for example: Vehicle cleaning, General hardware maintenance, Sensor wear replacement and recalibration, Customer support, etc. Even in a more secure, In a more automated world, We will still encounter unexpected situations, Many customers like to ask for help when they have problems. Can robots empathize and respond to human and unexpected situations? This is an important consideration for us, Because we want to develop a user experience that builds trust and understanding.

Driverless cars generate gigabytes of data per minute, All these data must be able to measure the ride quality, Track any critical events, And improve technology over time. Some data can be transmitted by mobile phone, But most of the data has to be stored in the car, This means that the hard disk must be replaced regularly. When the data leaves the vehicle, They will go directly to our multi-level data pipeline, We will process the data, For future playback, debugging, visualization, Notes and other analysis.

human-computer interaction:
The passengers in the unmanned vehicle chose the robot to handle the driving task. Inside the car, Our passengers can intuitively understand what the car sees and how it makes decisions. however, Outside the vehicle, The interaction between our vehicle and its environment is another matter. The departure of human drivers, It means that we no longer have the ability to make eye contact and express with the outside world, At least not in the same way. We know that the transition from driving to driverless will not happen overnight. We've designed a system to build trust in the future of autonomous driving. Human computer interaction is another important consideration.

This driverless marathon is a good learning opportunity. Deploy on a single unmanned vehicle“ Marathon” Let me们有机会了解如何使一整队无人车保持最佳性能.而在2018年,我们的目标就是:打造无人车队!




✪  高频面试考点

✪  行业项目经验

✪  简历修改完善

✪  面试注意事项