Home > News content

China's First Delivery Autopilot Unicorn: Fully Autonomous Parking, 1000-yuan Hardware Cost

via:博客园     time:2019/7/12 17:37:08     readed:61

Li Gen from the concave temple

Quotation Report Public Number QbitAI

This is not the first autonomous parking program.

But this is the first unmanned technology, the most difficult scenario to hold, and the real cost-effective solution.

The meaning is not limited to this. Under the unfavorable trend of the Tesla route, the global autonomous driving startups gossip more than business progress, unmanned vehicles "cold winter", delivery at this time is extraordinary.

Today's name:Momenta.

This Chinese self-driving unicorn company has just released its own parking products.Mpilot Parking.

This is a product, not to mention, look at the goods first.

Product first experience

Autonomous parking, as the name suggests, is to park the car in the whole process to let the car "autonomy".

At the 500-square-meter multi-storey office parking garage on the ground floor of the Momenta Suzhou headquarters, we experienced the whole process with the Lincoln MKZ, which has no obvious traces of self-driving modification:

Upon arrival at the parking lot, the driver gets off the bus and starts the autonomous parking:

data-copyright=0

Start to enter the ground from the ground, downhill, and the light is dimmed.

In the basement driving, the speed is about 5 kmhour, but the mixed scenes of people and vehicles are very common:

data-copyright=0

There are also other human drivers who are parking, and the drone chooses to stop and wait for it to complete:

data-copyright=0

Finally, arrive at the target parking space and rewind into the warehouse. If it is a non-office scenario, you can also specify or let the unmanned car “autonomy” to find free parking spaces:

data-copyright=0

Then switch the usage scene —— the unmanned vehicle is remotely summoned by the owner, and the parking is:

data-copyright=0

I encountered other vehicles again, but this time the unmanned vehicle detection decision can be bypassed, so I bypassed the —— similar scenes, different treatments, also reflected the side of the system's old drivers.

data-copyright=0

Finally arrived at the elevator, the owner got on the bus, the automatic driving system continued to help open the basement.

The whole process is zero-taken, and the owner can start and summon with one button remotely through the mobile phone.

It can be expected that once the plan is put into the front loading and delivered by the production car, the time for parking and storage will be saved, and many car park accidents can be greatly avoided.

And there is a timetable for excitement. Momenta revealed that the actual cooperation with multiple benchmark customers has started, and the product plan will be delivered through production cars from 2021 to 2022.

Automated driving product inspection standard

Although it is not the first to introduce autonomous parking products, Momenta's products are indeed different.

Is it different? It is reflected in the different aspects of the owner experience.

Different from the previous autonomous parking program —— no limitation scene, no expensive sensors, no even relying on intelligent parking lot modification.

Products can withstand three major standards of testing: available, easy to use, easy to use.

First fromAvailableSpeaking of the core test item is the cost.

An autonomous parking program, from the garage entrance to the parking space stalled and stalled, is not technically impossible today. However, once the laser radar sensor such as laser radar is needed, the mass production threshold will be greatly increased.

And when we say "available", it has to be affordable for everyone, any model can be used, no need to wait for years to wait for the standard of the car.

Momenta's plan is actually the result of this backward push.

The entire sensor system is equipped with a 4-way revolving fisheye camera, a front-view camera, and 12 ultrasonic radars, all of which are mature mass-produced sensors.

data-copyright=0

Although the cost data is not clearly published, but slightly familiar with the automotive industry supply chain, it is about the price of a mobile phone & mdash; & mdash; or domestic mobile phones.

If the cost of the sensor is more than one mobile phone, can the garage be handed in to the machine every day, is it considered usable?

Followed byEasy to useThe core test is the ability to respond to the scene.

In other words, any scene does not require the ability of the owner to take over at any time.

Different from the loop unmanned, the speed of the basement is relatively low, but there are still many extreme scenes.

data-copyright=0

For example, scenes where people and other vehicles are not operating properly, scenes where vehicles occupy roads or beyond normal parking spaces, and scenes where environmental changes are caused by multi-storey garages.

It will affect the normal driving of automatic driving. It also requires the autonomous parking scheme to accurately sense the type, size, distance and movement state of obstacles, as well as more accurate and accurate positioning capabilities.

data-copyright=0

FinallyEasy to use. The core requirement is to buy it out and not pick the venue.

The autonomous parking plan that has been released before, or the high requirements for the configuration of the vehicle-side smart sensing device, or the dependence on the intelligent transformation of the parking lot.

The most common is to use the sensor to modify the parking lot, and to reduce the difficulty of the smart end of the car through the way of road coordination, but both the cost and the modification time bring new problems.

This will hinder the true use of autonomous parking.

From the perspective of the car manufacturer and the owner, who wants to sell cars and new solutions, but also let users consider whether all parking lots can go?

Therefore, the field side does not solve the problem for one day, and the independent parking plan cannot be considered easy to use.

And since it is easy to get started, you have to "out of the box" & rdquo;, can no longer let users participate in the long-term "road test".

Therefore, this time Momenta also put its other core barrier capability & mdash;— high-precision map construction and positioning technology to the product plan.

Behind the technical details

The implementation of such a level of autonomous parking programs, the use of technology is certainly more than one.

The industry's familiar perception, decision making, control, and positioning all play a role.

But this Momenta programEasy to useThe most powerful nature isHigh-precision map constructionA breakthrough in terms.

Those who are familiar with Momenta's entrepreneurial history are no strangers to Momenta's high-resolution map scheme.

But now, “heavy work” is being converted into an advantage.

data-copyright=0

On the one hand is Momenta's proven low-cost automated mapping.

Compared to expensive lidar mapping, Momenta uses a vision-based approach to achieve low-cost automated mapping.

The mapping acquisition system uses a mature four-way revolving fisheye camera, consumer IMU and wheel speed sensors for a lower total cost.

In the process of mapping, the visual semantic features are extracted through deep learning methods, and SLAM technology is used to automatically generate semantic-based high-precision maps.

The entire system can be implemented in both the cloud and the car, without the involvement of personnel, with an accuracy of 10 cm. What is this concept? The general lane line is 20cm.

data-copyright=0

On the other hand, it is the low threshold for the construction: the production car can be completed by driving through the parking lot.

When the vehicle equipped with Mpilot Parking enters the uncovered area of ​​the map, the vehicle can self-learn through the parking lot, establish a map of the area, and upload it to the cloud for subsequent positioning based on the map. At the same time, other cars driving in this area can also enjoy the benefits.

This kindCrowdsourcingThe way is also the key to rapid scale to achieve low-cost mapping and updates.

No need for special map collection vehicles, support for mass production vehicles to build their own maps, and multiple vehicles to achieve rapid scale construction and update through crowdsourcing.

And as time goes by, elements added or disappeared in the basement can be updated in real time by crowd-carrying vehicles.

I have to say that it is indeed a way of dealing with the people's ocean and sea in the "infinite war".

data-copyright=0

Road to Momenta

It is worth mentioning that the core person behind the program: Xia Yan.

She is a partner at Momenta and a product development lead.

data-copyright=0

Although there are not many appearances in the industry, Xia Yan has long been famous in the academic field.

And Mpilot Parking is not the only product of the Momenta team.

They also built an autopilot solution for the Highway (high-speed), and will release the Mpilot Urban (city) scenario in the first quarter of next year.

What is commendable is that Momenta will be based on the same hardware system platform and sensing solution from high speed, parking to urban range.

This also means that there is no need to change the cost, it can provide automakers and owners with automatic driving ability in three scenarios to maximize the continuous automatic driving experience.

But this is not the complete route of this self-driving unicorn.

Xia Yan said that Momenta has set a clear goal since its inception, and hopes to realize the human-machine driving to completely unmanned driving through data iteration.

From someone to no one, the key is the real large-scale, automated circulation of data, forming a closed loop.

data-copyright=0

This puts high demands on the integration of technology, cost, safety and products.

In this program's landing, Xia Yan also specifically talked about the field-side intelligence, 5G and other technologies.

She believes that these technologies will have a huge impact on future autonomous driving, but they are now ready for redundancy in the future, but ultimately they will complete the core problem from the car end.

Therefore, we aim at mass production and delivery, and we also aim at mass production.

Xia Yan led the team to build a solution from the end of 2017, day and night, and constantly promote product iteration.

Now, Highway (high speed) and parking scenarios are completed, and pre-installation customer cooperation has begun.

In addition to the hardware sensing platform, this solution is also limited to models, which can maximize the mass production.

This is reminiscent of the latest thinking of Tesla CEO Mask. He said that once Tesla is fully automated, RoboTaxi will increase the price of the car and even sell the car.

Perhaps from the beginning, Musk was out of autopilot.

But on the other side of the globe, Momenta, a group of young AIers, drove from the beginning to autopilot, aiming to drive the brain in any car, so that everyone can experience the autopilot function first. .

First liberate part of the time, improve the driving experience, and finally realize the complete liberation of people from driving in the iterative circulation of data.

data-copyright=0

Momenta is equipped with a self-driving car, no obvious signs of modification

More importantly, Momenta stood up and talked with the product to give confidence to the outside world.

They set out because they believed in automatic driving.

But now, they will let more people believe because they saw it.

QubitQbitAI · headline signing author

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments