Intelligent obstacle avoidance device based on RISC-V
By adopting intelligent obstacle avoidance devices based on RISC-V in anti-collision systems, combined with millimeter-wave radar and micro machine learning technology, the problems of insufficient power consumption and real-time performance of existing systems are solved, and low power consumption and efficient intelligent obstacle avoidance capabilities are achieved.
Patent Information
- Application Number
- CN202510013224.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-06
- Publication Date
- 2025-05-06
AI Technical Summary
Existing anti-collision systems rely on high-cost hardware, with high power consumption, making it difficult to meet the strict requirements of real-time and power consumption.
Using an intelligent obstacle avoidance device based on RISC-V, combining millimeter wave radar technology, micro machine learning technology and RISC-V instruction set architecture, a low-power anti-collision system is built. The system includes a millimeter wave radar module, a micro machine learning processing unit and a processor based on the RISC-V instruction set. By optimizing the instruction set and algorithms, it realizes low-power real-time signal processing and machine learning inference.
It realizes efficient and intelligent obstacle avoidance capabilities in limited resources and low-power environments, improves the diversity and security of intelligent expansion of equipment, and significantly reduces power consumption, making it suitable for low-power end-side devices that operate for a long time.
Smart Images
Figure CN119942854A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of cycling safety technology, and in particular to an intelligent obstacle avoidance device based on RISC-V. Background Art
[0002] With the rapid development of autonomous driving and intelligent transportation, the anti-collision system of various vehicles has become a key technology to improve driving safety. Most of the current anti-collision systems rely on high-cost hardware such as proprietary processors and lidars, and have high power consumption, making it difficult to meet the strict requirements of real-time performance and power consumption. As an open source instruction set architecture, RISC-V has high customizability and low power consumption, which has attracted widespread attention in embedded applications. Millimeter-wave radar has become an important sensor for target detection due to its strong penetration and anti-interference ability. Micro machine learning technology can operate efficiently in resource-constrained environments and provide real-time intelligent analysis and decision support. Therefore, how to organically combine these three technologies to develop an efficient, low-power and real-time intelligent anti-collision system is an important direction of current technological development.
[0003] Most of the current collision avoidance systems rely on high-cost hardware such as proprietary processors and lidar, and have high power consumption and strict deployment requirements. At the same time, it is difficult to meet the strict requirements of real-time performance and power consumption. Summary of the invention
[0004] In order to solve the above technical problems, the present invention provides a RISC-V-based intelligent obstacle avoidance device, which is a collision avoidance system built using millimeter-wave radar technology, micro machine learning technology and RISC-V instruction set technology. It aims to improve the intelligent obstacle avoidance capability in limited and low-power scenarios, thereby improving the diversity and safety of the intelligent expansion of the equipment.
[0005] The technical solution of the present invention is:
[0006] A RISC-V-based intelligent obstacle avoidance device, comprising:
[0007] The millimeter wave radar module detects the position and speed of vehicles within the range through three-dimensional point cloud computing;
[0008] A micro machine learning processing unit that deploys the model on a RISC-V processor and uses the processor’s floating-point computing power and parallel processing capabilities for inference;
[0009] A processor based on the RISC-V instruction set to receive and process data from the millimeter-wave radar. The processor is optimized to perform signal processing tasks and machine learning algorithms with low power consumption.
[0010] The security warning and control module is responsible for receiving the reasoning and evaluation results from the micro machine learning module and automatically triggering corresponding protection measures according to the preset security policies.
[0011] Furthermore,
[0012] The millimeter-wave radar module is mounted on the front of the bicycle, with the angle and position optimized to maximize the detection range.
[0013] The millimeter wave radar module uses a second-order filtering mechanism to optimize the data:
[0014] (1) The first stage is to calculate the speed of vehicles within the range based on the 3D point cloud and posture data, and filter out targets with negative speeds, i.e. decelerating targets;
[0015] (2) The data from the first stage is used as input to detect the positions of the remaining targets, and the data of targets exceeding the safe distance is output and the user is reminded.
[0016] The output data of the millimeter-wave radar module is transmitted to the RISC-V processor through the serial port interface for preliminary processing.
[0017] Furthermore,
[0018] The micro machine learning processing unit is based on a deep neural network and is trained with task data. It has high accuracy and real-time performance and can perform reasoning operations even when processor resources are limited.
[0019] The micro machine learning processing unit determines whether the data is usable based on the data collected during the user's task and the evaluation of the task results after the task is completed;
[0020] If available, the user's riding location and speed information will be collected based on the millimeter-wave radar; combined with the fully-linked neural network local inference mechanism deployed in the RISC-V processor, the model will be locally upgraded.
[0021] The fully connected neural network local inference mechanism deploys the inference model to the RISC V processor after quantization and pruning, and also deploys a fully connected layer based on a convolutional neural network. The purpose is to perform local learning and model iteration on the position and speed data in this task in the fully connected layer according to the user's operation after each task is completed, so as to achieve the function of self-learning.
[0022] Furthermore,
[0023] The processor based on the RISC-V instruction set selects a RISC-V processor, configures an acceleration unit for processing radar data and machine learning reasoning, and adopts an optimized instruction set and algorithm; the instruction set is simplified or expanded according to the actual application scenario to improve processing efficiency.
[0024] Furthermore,
[0025] The safety warning and control module is integrated with the equipment's control system to ensure that the system can respond to potential dangerous situations in the shortest possible time.
[0026] The beneficial effects of the present invention are
[0027] Deep integration of millimeter-wave radar and RISC-V architecture: This invention deeply integrates the high-precision data of millimeter-wave radar with the processing power of RISC-V architecture to achieve real-time perception and processing of complex environments. Millimeter-wave radar provides high-resolution environmental data. In order to improve data quality, it filters out negative speed targets to optimize data flow. Then the RISC-V processor is responsible for real-time processing of this data, including denoising, feature extraction, etc., to provide high-quality input for subsequent machine learning reasoning.
[0028] Customized and optimized RISC-V processor: In response to the needs of millimeter-wave radar data processing and machine learning inference, the RISC-V processor has been specifically optimized, using a dedicated acceleration instruction set, retaining only peripheral interfaces with related functions, and removing redundant peripheral interfaces to improve data processing efficiency and real-time performance. Through modular design, the processor can be flexibly expanded or streamlined to adapt to different application scenarios, ensuring that the system maintains high performance while consuming very low power.
[0029] Lightweight micro machine learning model: A lightweight machine learning model is used, which is designed for resource-constrained environments. This model significantly reduces the use of computing resources while ensuring prediction accuracy. During the training process, the model is optimized for different driving scenarios to ensure that collision risks can be predicted quickly and accurately in a changing environment. At the same time, a fully connected neural network local inference mechanism is deployed on the processor side to achieve local iterative upgrades of the model.
[0030] Balance between real-time performance and low power consumption: Through the optimized design of the RISC-V processor and the efficient reasoning of the micro machine learning model, this system achieves a good balance between real-time performance and low power consumption. Compared with the traditional anti-collision system, the present invention significantly reduces power consumption while maintaining efficient detection and decision-making capabilities, making it more suitable for low-power end-side devices that run for a long time. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a schematic diagram of the working structure of the present invention. DETAILED DESCRIPTION
[0032] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0033] The present invention provides an intelligent obstacle avoidance device based on RISC-V, comprising:
[0034] Millimeter wave radar module
[0035] The radar is installed on the front of the bicycle, and the angle and position are optimized to maximize the detection range. The millimeter wave radar detects the position and speed of vehicles within the range through 3D point cloud computing. A second-order filtering mechanism is used to optimize the data: 1. The first stage is to calculate the speed of vehicles within the range based on 3D point cloud and posture data, and filter out targets with negative speed (deceleration). 2. The data from the first stage is used as input to detect the position of the remaining targets, output the target data that exceeds the safe distance, and alert the user. The output data of all radars is transmitted to the RISC-V processor through the serial port interface for preliminary processing.
[0036] A processor based on the RISC-V instruction set, a processor using the RISC-V open source architecture, is used to receive and process data from millimeter-wave radar. The processor has been customized and optimized to efficiently perform complex signal processing tasks and machine learning algorithms with low power consumption. The modular design of RISC-V allows the instruction set to be streamlined or expanded according to the actual application scenario, thereby improving processing efficiency.
[0037] A low-power RISC-V processor is selected and a dedicated acceleration unit is configured to process radar data and machine learning inference. The system uses optimized instruction sets and algorithms to ensure real-time response and low-power operation. The acceleration unit is used to process data from sensors (such as millimeter-wave radar, lidar, camera, etc.), mainly for signal preprocessing and feature extraction, and uses fast Fourier transform (FFT) to perform spectrum analysis on radar and other sensor signals. The accelerator integrates FFT and CFAR modules, which can quickly convert time domain data into frequency domain data and identify the distance and speed information between different objects. CFAR (Constant False Alarm Rate Detection) Optimization: This algorithm detects target objects in radar echo signals through adaptive thresholds. In the RISC-V acceleration unit, multiple threshold calculations are parallelized to reduce detection time and improve detection accuracy. Then the resolution and signal-to-noise ratio of the radar signal are improved through pulse compression algorithms. The signal processing accelerator can perform common filtering operations (such as Kalman filtering and Wiener filtering) to eliminate noise and improve the accuracy of target detection. Finally, combined with FFT, the radar signal is processed at multiple scales using a time-frequency analysis algorithm (such as wavelet transform) to identify the features of high-speed moving objects. Secondly, the acceleration unit accelerates parallel computing tasks such as matrix operations and vector operations in the process of machine learning model reasoning through a vector processor (VPU). The VPU supports vector extensions (RVV) in the RISC-V ISA and can process multiple sets of data in parallel, especially when processing radar signals and machine learning model reasoning. Vector operations can significantly increase the speed of operations such as matrix multiplication. For example, for convolution operations used for deep neural network reasoning, the VPU can parallelize the calculation of multiple convolution kernels to improve the computational efficiency of the CNN model. At the same time, the acceleration unit supports dynamic voltage and frequency scaling (DVFS), which can adjust the operating frequency and voltage according to the complexity of the current computing task. When the signal processing and machine learning reasoning tasks are light, the frequency is reduced to save power consumption, and the frequency is quickly increased when a potential collision risk is detected to ensure that the system can respond quickly.
[0038] The micro machine learning processing unit, the machine learning model is based on a deep neural network, and after a large amount of task data training, it has high accuracy and real-time, and can perform efficient reasoning operations under limited processor resources. The processing unit deploys the model on the RISC-V processor and uses the floating-point computing power and parallel processing power of the processor for efficient reasoning. According to the data collected during the user's task process, and then combined with the evaluation of the results of this task after the task is completed, it is determined whether the data is available. If available, the user's riding position and speed information will be collected based on the millimeter-wave radar; combined with the fully connected neural network local reasoning mechanism deployed in the RISC-V processor (this mechanism is the core technology for local model update. While the reasoning model is deployed to the RISC-V processor after quantization and pruning, a fully connected layer based on a convolutional neural network will also be deployed. The purpose is to perform local learning and model iteration in the fully connected layer after each task is completed according to the user's operation, so as to achieve the function of self-learning) to perform local model upgrades.
[0039] In order to improve processing efficiency and power consumption in the intelligent collision avoidance system, the instruction set and algorithm of the RISC-V acceleration unit have been optimized, including the RISC-V vector extension instruction set (RVV), which is a key instruction set for accelerating parallel computing tasks, especially in signal processing and machine learning reasoning tasks. RVV supports variable-length vector operations and significantly improves computing efficiency by processing multiple data elements in parallel. Vector Load / Store: Load / store multiple data elements with one instruction, reduce the number of memory accesses, reduce the pressure on memory bandwidth, and improve the overall processing speed. Vector Arithmetic: Contains basic arithmetic operations such as vector addition, subtraction, multiplication, and division. By processing multiple data points in parallel, it improves computing performance, especially in tasks such as convolution operations and matrix multiplication. Vectorized Convolution Instructions: In the process of machine learning reasoning, the convolution operation of convolutional neural networks is usually a computing bottleneck. RVV accelerates this process through a dedicated vector convolution instruction set, which can process multiple convolution kernels in parallel. In order to optimize the low power consumption and efficient reasoning performance of the intelligent collision avoidance system, the RISC-V acceleration unit replaces floating-point calculations (such as FP32) with fixed-point calculations (such as INT8 and INT16). Fixed-point operation optimization mainly includes: Fixed-point matrix multiplication: Matrix multiplication in the fully connected neural network layer can be accelerated by fixed-point operations in the INT8 format. Compared with traditional FP32 operations, fixed-point matrix multiplication occupies less memory bandwidth and operates faster. Fixed-point convolution: Used to accelerate the reasoning process of convolutional neural networks, fixed-point convolution operations can significantly reduce computing overhead and power consumption while maintaining high accuracy.
[0040] The safety warning and control module is responsible for receiving the inference evaluation results from the micro machine learning module and automatically triggering corresponding protective measures according to the preset safety strategy, such as sounding an alarm, automatic deceleration or emergency braking. This module is deeply integrated with the control system of the equipment to ensure that the system can respond to potential dangerous situations in the shortest time.
[0041] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.
Claims
1. An intelligent obstacle avoidance device based on RISC-V, characterized in that: include: The millimeter wave radar module detects the position and speed of vehicles within the range through three-dimensional point cloud computing; A micro machine learning processing unit that deploys the model on a RISC-V processor and uses the processor’s floating-point computing power and parallel processing capabilities for inference; A processor based on the RISC-V instruction set to receive and process data from the millimeter-wave radar. The processor is optimized to perform signal processing tasks and machine learning algorithms with low power consumption. The security warning and control module is responsible for receiving the reasoning and evaluation results from the micro machine learning module and automatically triggering corresponding protection measures according to the preset security policies.
2. The device according to claim 1, characterized in that The millimeter-wave radar module is mounted on the front of the bicycle, with the angle and position optimized to maximize the detection range.
3. The device according to claim 1 or 2, characterized in that: The millimeter wave radar module uses a second-order filtering mechanism to optimize the data: (1) The first stage is to calculate the speed of vehicles within the range based on the 3D point cloud and posture data, and filter out targets with negative speeds, i.e. decelerating targets; (2) The data from the first stage is used as input to detect the positions of the remaining targets, and the data of targets exceeding the safe distance is output and the user is reminded.
4. The device according to claim 3, characterized in that The output data of the millimeter-wave radar module is transmitted to the RISC-V processor through the serial port interface for preliminary processing.
5. The device according to claim 1, characterized in that The micro machine learning processing unit is based on a deep neural network and is trained with task data. It has high accuracy and real-time performance and can perform reasoning operations even when processor resources are limited.
6. The device according to claim 5, characterized in that The micro machine learning processing unit determines whether the data is usable based on the data collected during the user's task and the evaluation of the task results after the task is completed; If available, the user's riding location and speed information will be collected based on the millimeter-wave radar; combined with the fully-linked neural network local inference mechanism deployed in the RISC-V processor, the model will be locally upgraded.
7. The device according to claim 6, characterized in that The fully connected neural network local inference mechanism deploys the inference model to the RISC V processor after quantization and pruning, and also deploys a fully connected layer based on a convolutional neural network. The purpose is to perform local learning and model iteration on the position and speed data in this task in the fully connected layer according to the user's operation after each task is completed, so as to achieve the function of self-learning.
8. The device according to claim 1, characterized in that The processor based on the RISC-V instruction set selects a RISC-V processor, configures an acceleration unit for processing radar data and machine learning reasoning, and adopts an optimized instruction set and algorithm.
9. The device according to claim 8, characterized in that The instruction set is streamlined or expanded according to the actual application scenario to improve processing efficiency.
10. The device according to claim 1, characterized in that The safety warning and control module is integrated with the equipment's control system to ensure that the system can respond to potential dangerous situations in the shortest possible time.