Autopilot data solution: provide "fuel" for autopilot acceleration

faddiddn 2021-08-31 13:23:29 阅读数:89

autopilot data solution provide fuel

From new energy to driverless , Tesla has always been a leader in automotive technology , Musk has always believed that Tesla can solve the problem of automatic driving . On the third quarter earnings conference call , Musk just said full automatic driving (FSD) The feature is expected to be launched by the end of this year .

that , What is autopilot ?

Autopilot , It is a kind of unmanned ground vehicle , It has the transportation capacity of traditional vehicles . As an automated vehicle , Self driving vehicle can detect its environment and navigation without manual operation. .


Self driving vehicle can detect its environment and navigation without manual operation.

Automatic driving technology is divided into five levels

L0: The driver is in full control of the vehicle .

L1: The automatic system can sometimes assist the driver to complete certain driving tasks .

L2: The automatic system can complete some driving tasks , But drivers need to monitor the driving environment . At this stage , The error perception and judgment of the automatic system can be corrected by the driver at any time .

L3: The driver should always be ready to control the vehicle , During autopilot assist control , For example, it can be temporarily exempted from operation when following the car , But when the car detects the need for a driver , Will return immediately and let the driver take over his subsequent control .

L4: Advanced automatic driving , All operations are carried out by automatic driving system. , In the execution scenario (ODD) There is no illogical performance of the vehicle ( No mistakes ), There is no need for the owner's operation intervention .

L5: The automatic system can complete all driving tasks under any conditions .


Automatic driving level classification is made by NHTSA and SAE Proposed

What technologies are needed to realize automatic driving

Automatic driving can be said to be a comprehensive field involving many disciplines , Need identification technology 、 Decision making techniques 、 Location Technology 、 Computer interaction technology, overall coordination and cooperation to ensure driving safety .

Recognition technology

Like the human eye , Autopilot also owns “ eyes ”, Used to identify the surrounding vehicles 、 obstacle 、 Pedestrians and other conditions on the road .

Self driving “ eyes ” It usually consists of sensors , Including cameras 、 Laser radar 、 Millimeter wave radar , And infrared 、 Ultrasonic radar, etc .

Decision making techniques

The surrounding environment is obtained through eye recognition , Next, we should make full use of these data for understanding and analysis , Decide what to do next . To accomplish this task is “ The brain ”.

In order to make the right decision , Autonomous robots need some knowledge accumulation . There are ways to enrich the brain knowledge base : Expert rule and AI type .

Location Technology

Today, , Except for the mainstream GPS or GNSS( Global satellite navigation system ) Out of the way , There are also ways to lay electromagnetic induction lines on the highway to achieve positioning .

Special for automatic driving 3D Dynamic HD maps bring more possibilities for automatic driving . Because with HD maps , You can easily position yourself in the lane .

Human computer interaction technology

The ideal state of automatic driving should be , There is no need for people's intervention , Cars can take people wherever they want to go .

But unfortunately , At present, the automatic driving system can not do this. . Encounter scenes that autopilot can't control , The system will call the driver to take over its work .

A number of artificial intelligence technologies have been applied to the human-computer interaction scene of automatic driving : Driver identification by face recognition 、 Judge the driver's sleepiness through expression 、 Execute the instructions given by the driver by learning voice and gesture ……

At this time , The man-machine interface works . Its goal is to , Inform the driver to pay attention to the new situation as soon as possible in the most intuitive and convenient way 、 Execute driver command .


BMW New 5 Application of dynamic gesture recognition technology

Automatic driving data solution

With the continuous improvement of automatic driving technology , More and more automobile manufacturers 、 The system drives automatically towards altitude L4、 Fully autonomous driving L5 Advance quickly .

2020 year 7 World Artificial Intelligence Conference (WAIC) At the opening ceremony , Musk has said , Tesla has “ Very close to ” Realization L5 Class a autonomous driving technology , The development of the basic functions of this technology will be completed within this year .


Musk expressed confidence that it will be realized within one year L5 Level autopilot

As levels grow , The difficulty of making decisions by autopilot cars is also increasing. . Data is the basis for continuously enhancing the recognition and judgment ability of vehicles , It is also the key to the development of automatic driving technology .

The data hall is based on a deep understanding of intelligent driving scenarios , Through the box selection of massive raw data 、 extract 、 Classification and so on , Transfer the mixed and disordered data into intelligent driving professional data that can be recognized by machine learning , Assisted automatic driving technology can better perceive the actual road in complex and diverse environments 、 Vehicle location and obstacle information, etc , Real time perceived driving risk , Realize intelligent driving 、 Predetermined goals such as automatic parking .

Automatic driving technology needs large-scale data to train its automatic driving ability , Teach the car to understand the driving environment 、 Correct driving . Based on this , The data hall provides automatic driving data customization service .

Automatic driving data customization service capability

· Track ID: For specific target objects such as cars 、 Pedestrians 、 Track roadblocks, etc .

Example of vehicle tracking label

· 3D Point cloud annotation : Support 3D Single frame annotation 、2D-3D Mapping single frame dimensions 、3D Tracking labels 、2D-3D Joint tracking .


3D Point cloud annotation example

· traffic sign : For specific traffic signs such as traffic signs 、 Traffic light marking .


Example of traffic sign annotation

· Free space: Mark the driving area 、 Entity segmentation and road semantic segmentation .


Free space Examples

· Vehicle marking : Support vehicle labeling 、 vehicle 3D mark .


Example of vehicle annotation

· Road line annotation : Opposite lane line 、 Boundary line 、 Sidewalk line, etc .


Example sidewalk line dimensions

· Human body labeling : Support human body frame 、 Human skeleton key annotation 、 Driver behavior labeling 、 Interior environment voice marking, etc .


Example of human bone key annotation

· Driver behavior labeling : Support the driver to answer the phone during driving 、 yawn 、 Marking of dangerous driving behaviors such as line of sight deviation and fatigue driving behaviors .


Driver behavior annotation example

Data hall data customization service has its own unique advantages .

The first is the platform advantage . Global collection of resources through digital plus platform link , Plus the professional data processing base of data hall 、 Professional technical team support , It can quickly and continuously improve work efficiency . The cost of data acquisition and processing is low 、 High quality .

The second is technical advantages . Datahall has rich annotation tools , Overlay image 、 Text 、 video 、 Voice and other data types . Intelligent data processing supports the preprocessing of different types of data , It can quickly reduce labor cost , Increase of efficiency .

Last , Data security compliance of datahall . Data acquisition is authorized , The data hall stipulates that the data collection must obtain the power of attorney or electronic power of attorney signed by the collected person . And the data hall strictly abides by EU data protection regulations GDPR, Set up a full-time Data Protection Officer , Formulate data security management rules and strictly supervise their implementation , Also passed ISO27001 Information security certification , Ensure data security and compliance .

The emergence of automatic driving , Redefined urban and long-distance travel modes , It is also the interaction mode between people and technology in the vehicle , It brings new challenges .

How to design automatic driving system , Human computer interaction interface , Even between people , Interaction mode between systems , Will be a topic worthy of continuous exploration and excavation .