CN-122001886-A - Data flow driven robot low-delay communication system and method

CN122001886ACN 122001886 ACN122001886 ACN 122001886ACN-122001886-A

Abstract

The application relates to the technical field of robot software architecture and distributed systems, in particular to a low-delay communication system and method for a robot driven by a data stream. The system comprises a middleware scheduler, a sender computing node and a receiver computing node. The middleware scheduler is used for loading a predefined data stream topology description file to declare all permitted communication paths and carrying out admission check on a data stream sending request according to the file. The sender computing node is used for initiating a sending request, and after passing the verification, writing the data to be sent into the shared memory area in a unified memory layout format. The receiving side computing node is used for registering the receiving intention, and directly reading the written data in the shared memory area according to the instruction of the middleware scheduler after the receiving intention passes the verification. The application systematically solves the defects of the existing open robot middleware in the aspects of safety, instantaneity and resource utilization rate.

Inventors

ZHANG XIAODONG
Lv baorui
YANG ZIJIANG

Assignees

中国科学技术大学

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. The data flow driven robot low-delay communication system is characterized by comprising a middleware scheduler, a sender computing node and a receiver computing node; the middleware scheduler is used for loading a predefined robot data stream topology description file, wherein the data stream topology description file declares all permitted data stream paths in the system, and is also used for carrying out admission check on a data stream sending request from a computing node according to the data stream topology description file; the sender computing node is used for initiating a sending request to the middleware dispatcher, and writing data to be sent into a shared memory area in a unified memory layout format after passing the verification; The receiving side computing node is used for registering the receiving intention with the middleware dispatcher, and directly reading the data written in the shared memory area by the sending side computing node in a unified memory layout format according to the instruction from the middleware dispatcher after the receiving side computing node passes the verification.
2. The system according to claim 1, wherein: The unified memory layout format is an Apache Arrow type memory format.
3. The system according to claim 1, wherein: the predefined robot data flow topology description file declares, for each licensed data flow path, a corresponding sender computing node identification, destination computing node identification, and type of transmission data.
4. A method of data stream driven robotic low latency communication performed by a sender computing node in the system of claim 1, comprising: initiating a sending request to a middleware scheduler in the system; After the middleware scheduler performs admission check on the sending request according to a predefined data stream topology description file and passes the admission check, writing data to be sent into a shared memory area, wherein the data stream topology description file declares all permitted data stream paths in a system; Notifying the middleware scheduler.
5. A method of data stream driven robotic low latency communication performed by a middleware scheduler in a system according to claim 1, comprising: loading a predefined data stream topology description file that declares all licensed data stream paths in the system; Receiving a sending request from a sender computing node in the system, and performing admission verification on the sending request according to the data flow topology description file; after the verification is passed, the sender computing node is instructed to write the data to be sent into a shared memory area; And notifying the access information of the shared memory area to the permitted receiver computing node in the data stream topology description file.
6. The method of claim 5, wherein the instructing the sender computing node to write the data to be sent to the shared memory area comprises: And dynamically distributing the shared memory blocks with matched sizes for writing according to the metadata of the data to be transmitted.
7. The method of claim 5, wherein the admission check comprises: Any data flow requests not declared in the data flow topology description file are intercepted.
8. The method of claim 5, wherein the method further comprises: And immediately recovering the corresponding shared memory area after monitoring that all the permitted receiver computing nodes have completed data reading.
9. A method of data stream driven robotic low latency communication performed by a recipient computing node in the system of claim 1, comprising: Registering a reception intention with a middleware scheduler in the system; And directly reading the data written into the shared memory area by the sender computing node according to the instruction from the middleware scheduler after the middleware scheduler performs admission check on the receiving intention according to a predefined data flow topology description file and passes the admission check, wherein the data flow topology description file declares all the permitted data flow paths in the system.
10. The method of claim 9, wherein the direct reading comprises: and reading the data in the shared memory area through the memory pointer.

Description

Data flow driven robot low-delay communication system and method Technical Field The application relates to the technical field of robot software architecture and distributed systems, in particular to a low-delay communication system and method for a robot driven by a data stream. Background In the field of high-performance real-time computation such as robots and autopilots, middleware is used as a digital nervous system for connecting each perception, decision and control module, and the communication performance and the safety of the middleware are important. Currently, an open communication architecture is generally adopted by a mainstream robot middleware (such as ROS 2) in the industry, and the core of the open communication architecture is to implement data exchange between nodes based on a topic mode of Publish/Subscribe (Publish/Subscribe). However, this architecture, when dealing with high security, high real-time requirements, exposes the following interrelated fundamental drawbacks: First, open architecture lacks global admission control in terms of communication security and certainty. Any computing node can subscribe freely as long as knowing the topic name, so that the data flow of the core sensor such as a camera, a laser radar and the like is at risk of illegal interception or tampering, and the harsh requirements of the industry and the vehicle-mounted system on the logic isolation of the data cannot be met. Second, in terms of communication efficiency, the architecture must perform serialization and deserialization operations before and after data transmission. When processing large-size data such as images, point clouds and the like, the process can generate huge CPU computing power expenditure and remarkable communication delay, and becomes a key bottleneck of system instantaneity. Although the cross-language columnar memory format such as Apache Arrow provides a basis for eliminating serialization overhead and realizing zero-copy data exchange, the existing open architecture cannot be combined with a safe scheduling model with admission control, and the performance potential of the open architecture cannot be exerted. Finally, in terms of resource management, the matched shared memory transmission mostly adopts a pre-allocated buffer area with fixed size, so that the memory is idle and wasted when small data are transmitted, and the low-efficiency slicing and recombination are needed when large data are transmitted, and the memory fragments are easily caused by the hysteresis of memory recovery, so that the stability of long-term operation of the system is influenced. In summary, the existing open communication architecture mainly has three defects of potential safety hazard caused by open subscription, performance overhead caused by serialization/deserialization, and a stiff shared memory management mechanism. Disclosure of Invention The embodiment of the application provides a low-delay communication scheme of a data flow driven robot, which aims to solve the problems of potential safety hazards of data and leakage caused by lack of global admission control, huge communication delay and CPU (Central processing Unit) computing overhead caused by serialization/deserialization operation and low shared memory management rigidness and resource utilization rate caused by adoption of a fixed buffer zone in the prior art. The first aspect of the embodiment of the application provides a data flow driven robot low-delay communication system, which comprises a middleware dispatcher, a sender computing node and a receiver computing node; the middleware scheduler is used for loading a predefined robot data stream topology description file, wherein the data stream topology description file declares all permitted data stream paths in the system, and is also used for carrying out admission check on a data stream sending request from a computing node according to the data stream topology description file; the sender computing node is used for initiating a sending request to the middleware dispatcher, and writing data to be sent into a shared memory area in a unified memory layout format after passing the verification; The receiving side computing node is used for registering the receiving intention with the middleware dispatcher, and directly reading the data written in the shared memory area by the sending side computing node in a unified memory layout format according to the instruction from the middleware dispatcher after the receiving side computing node passes the verification. In some embodiments of the present application, the unified memory layout format is an Apache arow column memory format. In some embodiments of the present application, the predefined robot data flow topology description file declares, for each licensed data flow path, a corresponding sender computing node identification, destination computing node identification, and type of transmission data. A second aspect of an embod