DE102019219664A1 - Network arrangement for a vehicle and method for operating such a network arrangement - Google Patents

Abstract

The invention relates to a network arrangement (10a, 10b) for a vehicle comprising several network participants (1a, ..., 1n) which are designed to generate raw data, the network participants (1a, ..., 1n) designed to send the raw data are, further comprising a control system which is designed to receive data, several KI chips (2a, ..., 2n) are provided and each network participant (1a, ..., 1n) an KI chip (2a ,. ., 2n), each of the KI chips (2a, ..., 2n) being designed to receive the raw data generated by the network participants (1a, ..., 1n), each of the KI chips (2a , ..., 2n) is designed to process the raw data into processed data and to send the processed data to the control system. The invention also relates to a method for operating such a network arrangement.

Description

The invention relates to a network arrangement for a vehicle comprising a plurality of network participants which are designed to generate raw data, the network participants being designed to send the raw data, the network arrangement further comprising a control system which is designed to receive data. The invention also relates to a method for operating such a network arrangement.

Different bus systems are usually used in a vehicle to transmit data from different bus users. These have a high bandwidth; However, the large amount of information that arises during autonomous driving, for example, can lead to a data jam that endangers autonomous driving. Switches and other devices can only partially solve this problem and also lead to increased costs.

The DE 10 2013 223 704 A1 discloses a vehicle with a bus system which is designed as an Ethernet bus system and has at least one distribution device with a plurality of port connections to each of which a bus subscriber is connected, which transmits data to the distribution device via the assigned port connection during operation of the bus system and / or receives from the distribution facility.

It is an object of the invention to specify means for more secure operation in a vehicle network with various subscribers. Safe operation means high availability of the participants as a quality measure. As a result, with the high availability, robustness against impairments, including total failure, is achieved in one or more transmission channels that could lead to an error.

This object is achieved by a network arrangement having the features of claim 1 and a method having the features of claim 12.

The subclaims list further advantageous measures which can be suitably combined with one another in order to achieve further advantages.

The object is achieved by a network arrangement for a vehicle comprising a plurality of network subscribers designed to generate raw data, the network subscribers being designed to send the raw data, further comprising a control system designed to receive data, wherein a plurality of AI chips are provided and each network participant is assigned an KI chip, each of the KI chips being designed to receive the raw data generated by the network participants, and each of the KI chips for processing the raw data into processed data and for sending the processed data to the control system is designed.

The term “several” means at least two.

An AI chip can also be assigned to several network participants. AI stands for artificial intelligence.

AI chips are to be understood as chips that only master the necessary functions, i.e. cannot perform more complex operations. However, these AI chips are particularly powerful. Furthermore, the AI chips each have one or more AI processors which are able to operate an AI method, for example artificial neural networks. Such an artificial neural network is usually a deep neural network with several neurons as input, which receive information such as image parts or data from a radar / lidar sensor and forward it to several hidden layers between the input and the output. The neurons can be weighted differently.

The AI chips are also able to translate the raw data recorded by the network participants as input into, for example, a defined object recognition, such as vehicles, traffic lights or other road users, as processed data. Another processing of the raw data is also possible.

A control system preferably comprises several control devices.

The network arrangement according to the invention decentralizes the processing, in particular the evaluation of the raw data; Instead of in a central unit, for example in a central control unit, the evaluation is carried out decentrally in the AI chips and only the evaluated (processed) data is forwarded. This results in a reduction in data (size) while the information content remains the same. In this way, the processed data can be replicated multiple times and then forwarded via any vehicle interface without the risk of data congestion.

The network arrangement according to the invention thus improves traffic safety increased in the case of an autonomous mode of operation, since there is no longer a central processing node. The network arrangement according to the invention also avoids the use of complex interfaces with a high bandwidth such as MIPI / CSI, PCI Express, Ethernet. Furthermore, the network arrangement according to the invention reduces the system complexity and increases the reliability.

Each of the KI chips preferably comprises at least one KI processor, which is designed as a multi-core processor, also called an KI multi-core processor. A multi-core processor is a processor (CPU) that has several processor cores. Such multi-core processors provide a higher performance potential.

Each of the AI chips is preferably designed as an embedded system. These can be used in a simplified manner in vehicles.

Furthermore, the KI chip preferably comprises several KI processors. This can be several hundred AI processors. This speeds up the calculation process.

The control system preferably comprises at least one control device which is designed to receive data. The control device can be designed as a central control device or, for example, as a driver assistance control device. Furthermore, the control system preferably comprises a plurality of control devices, at least one of the AI chips being designed to transmit the processed data to at least two control devices.

In particular, several repeaters are provided, each KI chip being assigned a repeater, so that the processed data sent by the respective KI chip to the control system are amplified by the respective repeater. This increases the signal range while maintaining signal integrity. These repeaters, also called regenerators, amplify the signal that is sent out by the AI chips. These components are very inexpensive.

Furthermore, at least one LAN connection is preferably provided between at least one of the KI chips and the corresponding repeater, the one KI chip being designed to transmit the processed data to the corresponding repeater via the LAN connection.

Alternatively, at least one WLAN connection is provided between at least one of the KI chips and the corresponding repeater, the one KI chip being designed to transmit the processed data to the corresponding repeater via the WLAN connection.

Furthermore, at least one LAN connection is preferably provided, with at least one of the repeaters being designed to transmit the processed data to the at least one control device via the LAN connection. Alternatively, a WLAN connection is provided, with at least one of the repeaters being designed to transmit the processed data to the at least one control device via the WLAN connection.

In a further preferred embodiment, the network participants are designed as sensors and / or actuators. The sensors can be designed, for example, as an image camera, a time-of-flight camera, lidar or radar sensors, which are required for an autonomous mode of operation of the vehicle.

Furthermore, the AI chips are preferably designed to work on the basis of a neural network or a convolutional neural network or neuromorphic computing or spiking neural network.

• - Erzeugung von Rohdaten durch mehrere Netzwerkteilnehmer, und senden der Rohdaten durch die jeweiligen Netzwerkteilnehmer,
• - Empfangen der gesendeten Rohdaten durch einen dem jeweiligen Netzwerkteilnehmer zugeordneten KI-Chip,
• - Verarbeiten der Rohdaten zu verarbeiteten Daten durch den jeweiligen KI-Chip,
• - Senden der verarbeiteten Daten an das Steuerungssystem.

The object is also achieved by a method for operating a network arrangement for a vehicle with a control system comprising the steps:

• - Generation of raw data by several network participants, and sending the raw data by the respective network participants,
• - Receipt of the sent raw data by an AI chip assigned to the respective network participant,
• - Processing of the raw data into processed data by the respective AI chip,
• - Sending the processed data to the control system.

The advantages of the network arrangement according to the invention can also be transferred to the method. In particular, the method is suitable for being carried out on the network arrangement according to the invention.

Furthermore, the AI chips preferably work on the basis of a neural network or a convolutional neural network or neuromorphic computing or spiking neural network.

• 1: eine Netzwerkanordnung gemäß dem Stand der Technik,
• 2: eine erfindungsgemäße Netzwerkanordnung in einer ersten Ausgestaltung,
• 3: eine erfindungsgemäße Netzwerkanordnung in einer zweiten Ausgestaltung,
• 4: ein erfindungsgemäßes Verfahren.

Further features, properties and advantages of the present invention emerge from the following description with reference to the accompanying figures. They show schematically:

• 1 : a network arrangement according to the prior art,
• 2 : a network arrangement according to the invention in a first embodiment,
• 3 : a network arrangement according to the invention in a second embodiment,
• 4th : a method according to the invention.

Although the invention has been illustrated and described in more detail by the preferred exemplary embodiments, the invention is not restricted by the examples disclosed.

1 shows a network arrangement 100 according to the prior art.

The network arrangement 100 according to the prior art has a plurality of control units (ECUs) 101, 102, 103, of which the control units (ECU) 101, 102 are described in more detail.

The control device 101 has several port connections 104, 105, 106.

Furthermore, the network arrangement 100 has several sensors / actuators 107, 108, 109 which are connected to the port connections 104, 105, 106 of the control device 101 via a bus 110, for example an Ethernet bus and the corresponding bus lines. For this purpose, the sensors 107, 108, 109 can also have corresponding port connections 111, 112, 113.

The control device 102 also has port connections 114, 115, 116. At these port connections 114, 115, 116 further sensors / actuators 117, 118, 119 with corresponding port connections 120, 121, 122 are connected to the control device 102 via a further bus 123. This also applies analogously to the control device 103. There can also be more or fewer port connections / sensors / actuators, as well as, for example, Ethernet connections or other connections, etc.

The control device 101 is designed, for example, as a central control device of a vehicle. The control device 102 can be designed, for example, as a camera control device for controlling and processing camera data or, for example, as a driver assistance control device.

The sensors 107, 108, 109, 117, 118, 119 can be designed as CMOS (image) sensors, for example. The bus 110, 123 can be designed as an Ethernet bus or, for example, as a CAN bus. These buses 110, 123 function as long as the data volume is low relative to the nominal bandwidth.

However, a lot of data is generated by the sensors and actuators that have to be processed in the control units 101, 102, 103, for example, to ensure reliable autonomous driving. However, the probability of a data congestion increases proportionally with the amount of data to be sent. Above a utilization of 50%, a capacity overload can arise and, as a result, good efficiency of the transmission performance within the network arrangement 100 can be prevented. Especially with interfaces with a high bandwidth such as MIPI / CSI, PCI Express, Ethernet, etc., it is therefore very difficult to distribute the multitude of data that arise with autonomous driving / operating mode.

To prevent this, switches are used, for example, which are designed as a coupling element in computer networks and which connect network segments to one another. Other possibilities include, for example, the use of a PCI Express ("Peripheral Component Interconnect Express"), a standard for connecting peripheral devices. However, these components add to the cost of a network arrangement 100 and the level of complexity of the signal integrity.

2 shows an embodiment of a network arrangement according to the invention 10a a distributed hardware architecture for a vehicle. This has several network participants 1a, .., 1n on. The network participants 1a, .., 1n are designed to record different physical quantities of the vehicle or vehicle components based on raw data.

The network participants 1a, .., 1n can be designed as lidar / radar sensors or cameras, which generate signals as raw data, for example from the vehicle environment or the vehicle interior.

Instead of sensors, other electronic components that generate data, such as actuators, interior cameras or a vehicle-to-vehicle or vehicle-to-X communication module, can also be used.

The network participants 1a, .., 1n furthermore each have an interface for data transmission. The network arrangement according to the invention also includes 10a multiple AI chips 2a, ..., 2n . A Kl chip 2a ..., 2n can include several KI multi-core processors. A multi-core processor is a processor (CPU) that has several processor cores. Such multi-core processors provide a higher performance potential.

An AI chip 2a, ..., 2n can also be designed as an AI system-on-a-chip.

An AI system-on-a-chip is understood to mean an embedded system with at least one AI processor.

Such an AI chip only masters the necessary functions, but is particularly powerful. These AI chips are chips whose AI processors are able to operate artificial neural networks, for example. The artificial neural network is usually a deep neural network with several neurons as inputs, which receive information such as image parts or data from a radar / lidar sensor and pass it on to the several hidden layers between the input and the output.

The neurons can be weighted differently.

These AI chips 2a, ..., 2n are able to get through the network participants 1a, ..., 1n to translate captured raw data as input into a defined object recognition, such as vehicles, traffic lights or other road users, as output. It is also possible to process the raw data into other or additional information.

The AI chips 2a, ..., 2n can also include AI processors that are able to work on the basis of a convolutional neural network or neuromorphic computing or spiking neural network.

Thus, a training and an application on an AI chip 2a, ..., 2n enabling real-time learning directly from the environment without the need for a cloud.

AI chips 2a, ..., 2n however, cannot perform complex operations or operating systems, but convert raw data into information and perform algorithms such as pixel segmentation, object identification, etc.

In the AI chips 2a, ..., 2n the raw data are thus converted into processed data with meaningful information, such as other recognized objects, such as pedestrians, vehicles, traffic obstacles. Processing into other information is also possible.

The number of AI chips 2a, ..., 2n preferably corresponds to the number of network participants 1a, ..., 1n .

In addition, there are several control units 3a , 3b , 3c to receive the from the AI chips 2a, ..., 2n processed data available. More than three control units can be used 3a , 3b , 3c or less can be provided.

The processed data can be transmitted via vehicle-internal LAN connections such as FlexRay (Registered Trademark), MOST (Media Oriented Systems Transport) and D ² B (Domestic Digital Bus), CAN-Bus (Controller Area Network), CAN-FD (CAN Flexible Data rate) from the AI chips 2a, ..., 2n to the one or more control units 3a , 3b , 3c be transmitted.

The raw data received from the network participants 1a, .., 1n can also be generated via vehicle-internal LAN connections such as FlexRay (Registered Trademark), MOST (Media Oriented Systems Transport) and D ² B (Domestic Digital Bus), CAN-Bus (Controller Area Network), CAN-FD (CAN Flexible Data) Rate) from the network participants 1a, .. 1n to the AI chips 2a, .., 2n be transmitted.

The network participants can also 1a, .., 1n also for receiving data from control units 3a , 3b , 3c generated and sent and via a corresponding interface of the respective KI chip 2a, .., 2n be forwarded, be designed.

By the network arrangement according to the invention 10a a data reduction is possible in which only the processed data is sent to the control units 3a , 3b , 3c be transmitted. A data jam is thus avoided.

3 shows a further embodiment of a network arrangement according to the invention 10b .

This in turn includes several network participants 1a, .., 1n which in turn are designed to generate raw data.

Furthermore, the network arrangement 10b several AI chips 2a, .., 2n on, with each of the network participants 1a, .., 1 n an AI chip 2a, .., 2n assigned.

The AI chips 2a, .., 2n can be used in the same way as in 2 be trained.

There are also several control units 3a , 3b , 3c to the reception
the one from the AI chips 2a, ..., 2n processed data available.

There are also repeaters 4a ,, .., 4n intended. These repeaters 4a, .. 4n also regenerators called, amplify the signal coming from the AI chips 2a, .., 2n be sent out.

Every AI chip is included 2a, .., 2n a repeater 4a, .. 4n assigned. That way, those from the AI chips can 2a, .. 2n processed data generated from the raw data is replicated multiple times and via any vehicle interface, in particular even a wireless interface, to the various control units 3a , 3b , 3c be transmitted.

By means of the network arrangement according to the invention 10a , 10b is the evaluation of the network participants 1a, .., 1n , usually moved forward raw data processed in a control unit and transferred to many AI chips 2a, .., 2n distributed. Furthermore, only the useful information, that is, the processed data, is replicated multiple times and sent to the respective control device 3a , 3b , 3c Posted.

The data can be transmitted via any vehicle interface such as CAN, CAN-FD, Flexray or wirelessly.

Data can also be transmitted from the network participants 1a, .., 1n to the respective AI chip 2a , .. 2n wirelessly.

By the network arrangement according to the invention 10a , 10b the concentration of the processing of all processes within a central control unit is avoided.

By the network arrangement according to the invention 10a , 10b traffic safety is also increased, especially in the case of an autonomous mode of operation, since there is no longer a central processing node. By the network arrangement according to the invention 10a , 10b the use of complex interfaces with high bandwidth such as MIPI / CSI, PCI Express, Ethernet is avoided. Furthermore, the network arrangement according to the invention reduces 10a , 10b system complexity and increases reliability.

4th shows a method according to the invention.

In a first step S1, the raw data are generated by the sensors and the actuators, for example by recording the surroundings of the vehicle.

In a second step S2, the raw data generated are sent to a respective AI chip 2a, .., 2n ( 3 ) Posted.

In a third step S3, the raw data generated are from the respective AI chip 2a, ..., 2n ( 3 ) received and evaluated or processed by means of artificial intelligence, for example a neural network. For example, the detected objects such as pedestrians and other vehicles can be extracted from the sensor data as information.

In a fourth step S4, the processed data are transmitted via a repeater 4a, .., 4n to various control units 3a , 3b , 3c ( 3 ) Posted.

List of reference symbols

1a, .., 1n

Network participant

10a, 10b

Network arrangement

2a, .., 2n

AI chip

3a, 3b, 3c

Control unit

4a, .., 4n

Repeater

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102013223704 A1 [0003]

Claims (12)

Network arrangement (10a, 10b) for a vehicle comprising a plurality of network participants (1a, ..., 1n) which are designed to generate raw data, the network participants (1a, ..., 1n) being designed to send the raw data, furthermore comprising a control system which is designed to receive data, characterized in that several KI chips (2a, ..., 2n) are provided and each network participant (1a, ..., 1n) has an KI chip (2a, ..., 2n), each of the KI chips (2a, ..., 2n) being designed to receive the raw data generated by the network participants (1a, ..., 1n), each of the KI chips (2a, ..., 2n) is designed to process the raw data into processed data and to send the processed data to the control system. Network arrangement (10a, 10b) according to Claim 1 , characterized in that each of the KI chips (2a, ..., 2n) comprises at least one KI processor, which is designed as an KI multi-core processor. Network arrangement (10a, 10b) according to Claim 1 or 2 , characterized in that each of the KI chips (2a, ..., 2n) is designed as an embedded system. Network arrangement (10a, 10b) according to one of the preceding claims, characterized in that each of the KI chips (2a, ..., 2n) comprises a plurality of KI processors. Network arrangement (10a, 10b) according to one of the preceding claims, characterized in that the control system comprises at least one control device (3a, 3b, 3c) which is designed to receive data. Network arrangement (10a, 10b) according to Claim 5 , characterized in that the control system comprises several control devices (3a, 3b, 3c) and at least one of the KI chips (2a, ..., 2n) is designed to transmit the processed data to at least two control devices (3a, 3b, 3c) is. Network arrangement (10a, 10b) according to one of the preceding claims, characterized in that several repeaters (4a, ..., 4n) are provided and each KI chip (2a, ..., 2n) has a repeater (4a, .. ., 4n), so that the processed data sent by the respective KI chip (2a, ..., 2n) to the control system are amplified by the respective repeater (4a, ..., 4n). Network arrangement (10a, 10b) according to Claim 7 , characterized in that at least one LAN connection is provided between at least one of the KI chips (2a, ..., 2n) and the corresponding repeater (4a, ..., 4n) and the one KI chip (2a, ..., 2n) is designed to transmit the processed data via the LAN connection to the corresponding repeater (4a, ..., 4n). Network arrangement (10a, 10b) according to Claim 7 , characterized in that at least one WLAN connection is provided between at least one of the KI chips (2a, ..., 2n) and the corresponding repeater (4a, ..., 4n) and the one KI chip (2a, ..., 2n) is designed to transmit the processed data to the corresponding repeater (4a, ..., 4n) via the WLAN connection. Network arrangement (10a, 10b) according to one of the preceding claims, characterized in that the network participants (1a, ..., 1n) are designed as sensors and / or actuators. Network arrangement (10a, 10b) according to one of the preceding claims, characterized in that the AI chips (2a, ..., 2n) are designed to be based on a neural network or a convolutional neural network or neuromorphic computing or spiking neural Network to work. Method for operating a network arrangement (10a, 10b) for a vehicle with a control system, characterized by the following steps: - Generation of raw data by several network participants (1a, ..., 1n) and the sending of the raw data by the respective network participants (1a,. .., 1n), - Receipt of the sent raw data by an KI chip (2a, ..., 2n) assigned to the respective network subscriber (1a, ..., 1n), - Processing of the raw data into processed data by the respective KI -Chip (2a, ..., 2n), - Sending the processed data to the control system.

Images