FPGA-based laser radar beam scanning path optimization and energy efficiency management method

Abstract

The invention relates to the technical field of laser radars, and discloses a laser radar beam scanning path optimization and energy efficiency management method based on an FPGA, and the method comprises the following steps: S1, collecting the point cloud data and obstacle distribution information of a target region in real time through an environment sensing module; s2, generating a dynamic scanning path based on space partition, and generating a self-adaptive spiral scanning path for avoiding obstacles; s3, executing multi-parameter collaborative energy efficiency adjustment, and dynamically matching the laser transmitting power, the receiver gain and the MEMS driving voltage; and S4, synchronously outputting the optimized scanning path and the energy efficiency instruction to a laser driving circuit and a light beam deflection controller through a hierarchical scheduling module of the FPGA. A detection area is divided into priority grids through an FPGA, a self-adaptive spiral scanning path is generated, a hardware-level sudden obstacle avoidance algorithm is utilized, the response time is shortened to be smaller than or equal to 5 ms, and the integrity of point cloud data is ensured.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and specifically to a laser radar beam scanning path optimization and energy efficiency management method based on FPGA. Background Art

[0002] As a core sensor for three-dimensional environmental perception, lidar is widely used in autonomous driving, robotic navigation, and industrial inspection. Its technical principle is to generate point cloud data by emitting a laser beam and receiving the reflected signal, thereby constructing a spatial model of the environment. Currently, mainstream scanning solutions fall into two categories: mechanical rotary scanning and solid-state MEMS micromirror scanning.

[0003] In the existing technology, lidar usually adopts a fixed mode (such as raster scanning or uniform circular scanning) for environmental detection. This method does not combine real-time obstacle distribution with historical scanning data, resulting in low scanning efficiency: on the one hand, the scanning density of high-value areas (such as close distances or areas with high information entropy) is insufficient; on the other hand, redundant scanning is performed on scanned or low-value areas, resulting in a waste of resources.

[0004] The inventors of this application discovered that the above-mentioned existing technologies have problems: the traditional fixed scanning mode and discrete energy efficiency adjustment mechanism result in low scanning efficiency and insufficient point cloud integrity of the lidar in dynamic scenes. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an FPGA-based lidar beam scanning path optimization and energy efficiency management method, which solves the problems of low scanning efficiency and insufficient point cloud integrity of lidar in dynamic scenes caused by the traditional fixed scanning mode and discrete energy efficiency adjustment mechanism.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a laser radar beam scanning path optimization and energy efficiency management method based on FPGA, comprising the following steps:

[0007] S1. Collect point cloud data and obstacle distribution information of the target area in real time through the environmental perception module;

[0008] S2. Dynamic scanning path generation based on spatial partitioning: the target area is divided into priority grids within the FPGA. The historical scanning data and real-time obstacle positions are combined to calculate the scanning value weight of each grid cell and generate an adaptive spiral scanning path that avoids obstacles.

[0009] S3, perform multi-parameter coordinated energy efficiency adjustment, dynamically matching laser transmission power, receiver gain and MEMS drive voltage according to the steering angular velocity and laser ranging frequency of the current scanning path;

[0010] S4. Through the hierarchical scheduling module of FPGA, the optimized scanning path and energy efficiency instructions are synchronously output to the laser driving circuit and beam deflection controller.

[0011] By adopting the above technical solution, the environmental perception module collects the target area point cloud data and obstacle distribution information in real time. The target area is divided into priority grids within the FPGA, and the scanning value weight of each grid unit is calculated by combining historical scanning data and real-time obstacle positions. An adaptive spiral scanning path is generated to avoid obstacles. At the same time, the laser transmission power, receiver gain and MEMS drive voltage are dynamically matched according to the steering angular velocity and laser ranging frequency of the current scanning path to perform multi-parameter coordinated energy efficiency adjustment. Finally, the optimized scanning path and energy efficiency instructions are synchronously output to the laser drive circuit and beam deflection controller through the FPGA's hierarchical scheduling module, effectively solving the problems of low scanning efficiency and insufficient point cloud integrity of lidar in dynamic scenes caused by traditional fixed scanning modes and discrete energy efficiency adjustment mechanisms.

[0012] Preferably, the adaptive spiral scanning path generation in step S2 includes: establishing a spatial cost function for the priority grid

[0013] C=α·D+β·I

[0014] Where D is the Euclidean distance between the grid and the radar, I is the information entropy attenuation coefficient of the historical scan data, and α and β are weighting factors dynamically adjusted by the FPGA;

[0015] Taking the minimum value of the cost function as the scanning starting point, a spiral path with a continuously variable radius is generated, and a real-time avoidance interpolation algorithm for sudden obstacles is embedded.

[0016] Preferably, the dynamic energy efficiency adjustment in step S3 includes:

[0017] The energy efficiency index of the path steering angular acceleration a and the laser emission frequency f is calculated using the power consumption model built into the FPGA.

[0018] E=k1·a 2 / f+k2·T

[0019] Where k1 and k2 are calibration coefficients, and T is the real-time data of the temperature sensor;

[0020] When E exceeds the preset threshold, the laser power is gradually reduced to a safe threshold, and the receiver gain is increased proportionally to maintain the signal-to-noise ratio.

[0021] Preferably, the FPGA hierarchical scheduling module includes:

[0022] Path calculation layer: uses a parallel pipeline structure to perform space partitioning and path generation;

[0023] Power consumption decision layer: An integrated lookup table stores the optimal power consumption configuration for different steering angle-laser frequency combinations;

[0024] Hardware driver layer: Scanning instructions and energy efficiency instructions are alternately output to the laser and MEMS driver through a double buffer mechanism.

[0025] Preferably, the sudden obstacle avoidance algorithm is:

[0026] A monitoring window is embedded in the scanning path. When a new obstacle is detected in the real-time point cloud, the current scanning task is terminated.

[0027] The sub-path replanning circuit is activated in the FPGA to generate a local B-spline curve alternative path with the obstacle contour as the constraint boundary and compensate for the interrupted point cloud data.

[0028] Preferably, the power consumption model is generated by offline training:

[0029] Collect current sensor data under different scanning modes and build a power consumption prediction neural network in the FPGA with angular acceleration, laser frequency, and ambient temperature as input;

[0030] Solidify the neural network weights as a multiplier-accumulator array in hardware logic circuits.

[0031] Preferably, the method further includes: starting an energy efficiency sleep strategy during the idle phase of the laser radar:

[0032] When no moving target is detected for N consecutive scanning cycles, it switches to low-power monitoring mode:

[0033] The scanning path is reduced to the preset key area, the laser emission power is reduced to the sleep threshold, and the MEMS actuator is switched to stepping mode operation.

[0034] Preferably, the hardware driver layer includes a reconfigurable logic unit:

[0035] Dynamically configure the duty cycle accuracy of laser pulse encoding to 1ns level;

[0036] The 12-bit high-precision PWM module generates an adaptive MEMS drive waveform to reduce steering mechanical losses.

[0037] Preferably, data compression is used for transmission between the environment perception module and the FPGA:

[0038] The point cloud data is filtered and differentially encoded by the voxel grid and then transmitted to the FPGA;

[0039] The obstacle outline is extracted by the edge detection coprocessor as a set of key point coordinates and written directly to the FPGA memory through the DMA channel.

[0040] Preferably, a laser radar system comprises:

[0041] FPGA chip, multi-line laser emission array, avalanche photodiode receiving array, two-dimensional MEMS micromirror, and temperature-current sensor group;

[0042] The FPGA chip integrates a scan path optimization pipeline, an energy efficiency decision state machine, and a multi-channel PWM hardware controller.

[0043] The present invention provides a method for optimizing the laser radar beam scanning path and managing energy efficiency based on FPGA. It has the following beneficial effects:

[0044] 1. The present invention divides the detection area into priority grids through FPGA, dynamically calculates the spatial cost function by combining the real-time obstacle position and historical scanning information entropy, generates an adaptive spiral scanning path, significantly improves scanning efficiency, and uses a hardware-level sudden obstacle avoidance algorithm to shorten the response time to ≤5ms, ensuring the integrity of point cloud data.

[0045] 2. The FPGA of the present invention has a built-in power consumption prediction neural network, which makes real-time decisions based on the energy efficiency index E: when E exceeds the threshold, the laser power is gradually reduced and the receiving gain is simultaneously increased to maintain the signal-to-noise ratio. In combination with the sleep strategy, static power consumption is reduced.

[0046] 3. In terms of FPGA hierarchical scheduling and system integration, the present invention adopts a three-level pipeline architecture consisting of a path calculation layer, a power consumption decision layer, and a hardware driver layer. It directly writes to the FPGA memory through a DMA channel. During peak traffic hours, the point cloud processing delay is ≤10ms, the system power consumption fluctuation is ≤±5%, and the accuracy of abnormal event detection reaches 99.8%.

[0047] 4. When no moving target is detected for N consecutive scanning cycles, the present invention switches to a low-power monitoring mode, reduces the scanning path to a preset key area, reduces the laser emission power to the sleep threshold, and switches the MEMS driver to stepping mode. This significantly reduces power consumption during inactive periods, and once a target is detected, the scanning path and energy efficiency configuration can be rebuilt in milliseconds. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the FPGA-based laser radar beam scanning path optimization and energy efficiency management method of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Please see the attached Figure 1 , an embodiment of the present invention provides a laser radar beam scanning path optimization and energy efficiency management method based on FPGA, comprising the following steps:

[0051] S1. Collect point cloud data and obstacle distribution information of the target area in real time through the environmental perception module;

[0052] S2. Dynamic scanning path generation based on spatial partitioning: the target area is divided into priority grids within the FPGA. The historical scanning data and real-time obstacle positions are combined to calculate the scanning value weight of each grid cell and generate an adaptive spiral scanning path that avoids obstacles.

[0053] S3, perform multi-parameter coordinated energy efficiency adjustment, dynamically matching laser transmission power, receiver gain and MEMS drive voltage according to the steering angular velocity and laser ranging frequency of the current scanning path;

[0054] S4. Through the hierarchical scheduling module of FPGA, the optimized scanning path and energy efficiency instructions are synchronously output to the laser driving circuit and beam deflection controller.

[0055] Specifically, first, the environmental perception module collects point cloud data and obstacle distribution information of the target area in real time. The point cloud data is transmitted to the FPGA after voxel grid filtering and differential coding compression. At the same time, the coordinate set of the key points of the obstacle contour is extracted through the edge detection coprocessor and written directly to the FPGA memory through the DMA channel.

[0056] Then, a spatial partitioning strategy is adopted in the FPGA to divide the target area into priority grids. The grid scanning value weight is calculated by combining the information entropy decay coefficient I of the historical scanning data and the real-time obstacle position. The spatial cost function C = α·D + β·(1 / I) is constructed (D is the Euclidean distance between the grid and the radar, α and β are dynamic weighting factors). The minimum point of the function is used as the starting point of the spiral path to generate a continuously variable radius adaptive scanning trajectory. This trajectory is embedded in the real-time monitoring window. When the point cloud detects a sudden obstacle, the sub-path replanning circuit is activated. The obstacle contour is used as a constraint to generate a local B-spline curve and compensate for point cloud interruption. At the same time, multi-parameter collaborative energy efficiency adjustment is performed. The energy efficiency index E = k1·a of angular acceleration a and laser frequency f is calculated through the built-in power consumption model of FPGA. 2 f+k2·T (T is the real-time temperature, k1 and k2 are calibration coefficients). When E exceeds the threshold, the laser power is gradually reduced to the safety threshold and the receiver gain is increased proportionally.

[0057] Finally, the instructions are output synchronously through the three-layer FPGA scheduling module: the path calculation layer performs grid partitioning and path generation in a pipeline structure, the power consumption decision layer matches the optimal power consumption configuration based on a lookup table, and the hardware driver layer alternately outputs control signals through a double buffering mechanism. Its reconfigurable logic unit achieves 1ns-level laser duty cycle precision control and uses 12-bit high-precision PWM to generate adaptive MEMS drive waveforms to reduce mechanical losses.

[0058] The adaptive spiral scanning path generation in step S2 includes: establishing a spatial cost function for the priority grid

[0059] C=α·D+β·I

[0060] Where D is the Euclidean distance between the grid and the radar, I is the information entropy attenuation coefficient of the historical scan data, and α and β are weighting factors dynamically adjusted by the FPGA;

[0061] Taking the minimum value of the cost function as the scanning starting point, a spiral path with a continuously variable radius is generated, and a real-time avoidance interpolation algorithm for sudden obstacles is embedded;

[0062] The sudden obstacle avoidance algorithm is:

[0063] A monitoring window is embedded in the scanning path. When a new obstacle is detected in the real-time point cloud, the current scanning task is terminated.

[0064] The sub-path replanning circuit is activated in the FPGA to generate a local B-spline curve alternative path with the obstacle contour as the constraint boundary and compensate for the interrupted point cloud data.

[0065] Specifically, in the process of generating an adaptive spiral scanning path, the three-dimensional space segmentation operation is dynamically performed through the FPGA programmable logic, and a priority mapping matrix is ​​constructed based on the cubic grid unit. Based on the Euclidean distance D between each grid and the lidar body and the historical scanning information entropy attenuation coefficient I (the coefficient I is dynamically attenuated by the information entropy of the point cloud in the previous N cycles according to the model I = I0·e^(-λt)), the real-time updated weighting factors α and β are configured to construct the spatial cost function: C = α·D+β·(1 / I). The grid where the global minimum value of the cost function is located is selected as the scanning reference point, and the MEMS galvanometer is driven to generate an exponential spiral path with a continuously variable radius.

[0066] A hardware status monitoring window is embedded in the scanning path. When the point cloud coprocessor detects a new obstacle outline within a 1ms cycle, the current scanning timing chain is immediately interrupted, the beam deflector DAC output is frozen, and the B-spline path replanning circuit embedded in the FPGA is activated. The obstacle envelope point set is used as the constraint boundary, and a local avoidance path is generated through cubic B-spline interpolation (the path curvature is continuous and C 2A point cloud interpolation compensator is used to reconstruct the spatial trajectory during the interruption period, and an incremental coordinate counter is used to fill in the missing point cloud coordinates to ensure the spatiotemporal continuity of the point cloud data.

[0067] The dynamic energy efficiency adjustment in step S3 includes:

[0068] The energy efficiency index of the path steering angular acceleration a and the laser emission frequency f is calculated using the power consumption model built into the FPGA.

[0069] E=k1·a 2 / f+k2·T

[0070] Where k1 and k2 are calibration coefficients, and T is the real-time data of the temperature sensor;

[0071] When E exceeds the preset threshold, the laser power is gradually reduced to a safe threshold, and the receiver gain is increased proportionally to maintain the signal-to-noise ratio.

[0072] The power consumption model is generated through offline training:

[0073] Collect current sensor data under different scanning modes and build a power consumption prediction neural network in the FPGA with angular acceleration, laser frequency, and ambient temperature as input;

[0074] Solidify the neural network weights as a multiplier-accumulator array in hardware logic circuits.

[0075] Specifically, the dynamic energy efficiency adjustment module operates in real time through the power consumption prediction circuit integrated in the FPGA chip. In the online energy efficiency evaluation stage, the power consumption model is based on the calibration coefficient library (k,k z ) and the real-time data I of the temperature sensor, calculate the energy efficiency index of the current scanning path steering angular acceleration a and the laser emission frequency f:

[0076]

[0077] During the adaptive power adjustment phase, when E exceeds the preset safety threshold, the hierarchical power control unit sequentially performs the following operations: using the DAC controller to reduce the laser transmit power in 0.5W steps until it reaches the safety threshold; simultaneously, the receiver's automatic gain compensation circuit is activated to increase the gain by 1.2 times the power reduction ratio to maintain a signal-to-noise ratio of 15dB or higher; and a power consumption model hardware generation mechanism is implemented. A multi-source data acquisition chain consisting of a current sensor, a temperature sensor, and a motion status register (sampling rate of 1kHz) is constructed in the FPGA to collect power consumption data under different operating conditions offline.

[0078] A three-layer neuron computing array (input layer: angular acceleration / laser frequency / ambient temperature; output layer: dynamic current value) is implemented through on-chip programmable logic. After training, the weight matrix is ​​solidified into a parallel multiplier-accumulator (MAC) hardware structure to achieve microsecond-level power consumption prediction response.

[0079] The FPGA hierarchical scheduling module includes:

[0080] Path calculation layer: uses a parallel pipeline structure to perform space partitioning and path generation;

[0081] Power consumption decision layer: An integrated lookup table stores the optimal power consumption configuration for different steering angle-laser frequency combinations;

[0082] Hardware driver layer: alternately outputs scanning instructions and energy efficiency instructions to the laser and MEMS driver through a double buffer mechanism;

[0083] The hardware driver layer contains reconfigurable logic units:

[0084] Dynamically configure the duty cycle accuracy of laser pulse encoding to 1ns level;

[0085] The 12-bit high-precision PWM module generates an adaptive MEMS drive waveform to reduce steering mechanical losses.

[0086] Specifically, the FPGA hierarchical scheduling module adopts a three-level hardware logic pipeline architecture:

[0087] The path calculation layer implements a parallel space partitioning engine and path generation core through a reconfigurable logic array:

[0088] The spatial partitioning engine performs the fast cube mesh partitioning (clock cycle ≤ 200ns) of claim 1, and the path generation core synchronously calculates the spiral path parameter equation γ(θ) and the obstacle avoidance interpolation points.

[0089] The power consumption decision layer integrates an energy efficiency optimization decision matrix consisting of a 16-bit wide ROM lookup table (LUT): The LUT pre-stores the optimal power consumption configuration for the combination of steering angular acceleration α and laser frequency f. The real-time index key value is the (α, f, T) triplet (T is the quantized value of the temperature sensor), and outputs the laser power-receiving gain ratio pair to the hardware driver layer (response delay ≤ 5μs).

[0090] The hardware driver layer uses a double-buffered register set to implement alternating output of drive instructions: the first buffer outputs the scan angle coordinates to the MEMS driver via a 32-bit DMA channel (DAC refresh rate ≥ 1MHz)

[0091] ,The second buffer sends the energy efficiency control word to the laser modulator (including the power level and gain code) through the SPI bus. The two levels of buffer are switched during the vertical blanking period to ensure the phase synchronization of the control instructions.

[0092] Also includes: starting the energy-efficient sleep strategy during the lidar idle phase:

[0093] When no moving target is detected for N consecutive scanning cycles, it switches to low-power monitoring mode:

[0094] The scanning path is reduced to the preset key area, the laser emission power is reduced to the sleep threshold, and the MEMS actuator is switched to stepping mode operation.

[0095] Specifically, during the inactive period of the lidar, the hierarchical scheduling module automatically activates the low-power sleep strategy. When the mobile target detection logic integrated in the path calculation layer fails to identify a mobile target for N consecutive complete scanning cycles (typical value N = 5, configurable) based on real-time point cloud analysis, the system automatically switches to low-power monitoring mode.

[0096] The original full-scene dynamic scanning path is reduced to only cover the preset core high-priority monitoring areas (such as entrances, passages and other key areas, the area coordinates are pre-stored in the FPGA memory), and the working power of the laser emission module is reduced in real time to the preset sleep threshold power (typical value is 10%-15% of the rated working power, which is required to ensure the minimum detection capability), significantly reducing the active emission energy consumption.

[0097] The MEMS drive mode in the beam deflection controller is switched from continuous resonant scanning to an intermittent, low-duty-cycle stepping mode. In this mode, the micromirror performs precise, small step-by-step rotations only when needed to detect critical areas, significantly reducing the driver's dynamic power consumption and mechanical wear. During low-power monitoring mode, the system continuously scans critical areas. If a moving target is redetected or an external wake-up signal is triggered within any cycle, the hierarchical scheduling module immediately terminates sleep mode and, based on the current obstacle distribution information, reconstructs and executes the optimal adaptive scanning path and energy efficiency configuration within a few milliseconds, resuming normal operation.

[0098] Data compression is used for transmission between the environment perception module and the FPGA:

[0099] The point cloud data is filtered and differentially encoded by the voxel grid and then transmitted to the FPGA;

[0100] The obstacle outline is extracted by the edge detection coprocessor as a set of key point coordinates and written directly to the FPGA memory through the DMA channel.

[0101] Specifically, to achieve low-latency and high-bandwidth data transmission, a targeted data compression and optimized transmission mechanism is implemented between the environmental perception module and the FPGA executing the core logic. The raw dense point cloud data is first processed through voxel grid downsampling and filtering. Points that fall within the same grid are aggregated using a configurable spatial resolution (e.g., a 5cmx5cmx5cm cube grid), significantly reducing data density. Subsequently, differential encoding technology is applied to the filtered sparse point cloud, further compressing the data by recording the coordinate differences between consecutive points in 3D space rather than their absolute coordinate values.

[0102] The edge detection coprocessor integrated in the environmental perception module performs real-time analysis of raw or pre-processed perception data (such as images or point clouds), accurately extracting the outline features of obstacles (rather than complete patch data). The processor efficiently represents this outline information as a set of keypoint coordinates and compresses it. The compressed point cloud data stream and lightweight obstacle keypoint coordinate set are both written directly to the target buffer via a direct memory access channel, bypassing the FPGA's general-purpose processor core. This mechanism maximizes transmission efficiency, minimizes transmission latency and CPU intervention overhead, and ensures that the FPGA can immediately obtain the environmental information required for processing.

[0103] A laser radar system, comprising:

[0104] FPGA chip, multi-line laser emission array, avalanche photodiode receiving array, two-dimensional MEMS micromirror, and temperature-current sensor group;

[0105] The FPGA chip integrates a scan path optimization pipeline, an energy efficiency decision state machine, and a multi-channel PWM hardware controller.

[0106] The following is an introduction based on the embodiments

[0107] Example 1

[0108] Adaptive spiral scanning path optimization method based on space partitioning

[0109] Implementation scenario: Applicable to vehicle-mounted lidar in intelligent driving scenarios, which requires real-time scanning of obstacles (such as vehicles, pedestrians, and guardrails) in dynamic traffic environments.

[0110] Implementation steps

[0111] Environmental data collection and preprocessing

[0112] The environmental perception module uses a 16-line laser array to collect point cloud data of the target area. After voxel grid filtering (resolution 5 cm × 5 cm × 5 cm) and differential encoding compression, it is transferred to the FPGA via a DMA channel. The edge detection coprocessor extracts key points of the obstacle outline (such as the coordinates of the corner points of the car's outline) and writes them directly to the FPGA memory.

[0113] Priority grid division and cost function calculation

[0114] The FPGA divides the 50m × 30m × 10m space ahead into a 1m × 1m × 1m cube grid and constructs a priority mapping matrix. For each grid, the Euclidean distance D (straight-line distance to the radar) and the historical scan information entropy decay coefficient I (dynamically decayed according to I = I0·e^(-λt), with λ = 0.1 / s) are calculated. A pipelined architecture is then used to parallelize the calculation of the spatial cost function C = 0.6·D + 0.4·(1 / I).

[0115] Spiral path generation and obstacle avoidance

[0116] The minimum point of the cost function is selected as the scanning starting point, driving the 2D MEMS micromirror to generate a variable-radius spiral path (initial radius 0.5m, radius growth rate 0.2m / turn). A 1ms monitoring window is embedded in the path. When a sudden obstacle (such as a pedestrian crossing the road) is detected, the B-spline replanning circuit is activated, generating a local avoidance path based on the obstacle outline as a constraint, and compensating for missing data through point cloud interpolation.

[0117] Command output and hardware driver

[0118] The path calculation layer outputs scanning instructions through a double buffering mechanism. The hardware driver layer configures the laser pulse duty cycle to 1ns accuracy and generates a 12-bit PWM waveform to drive the MEMS micromirror to reduce steering loss.

[0119] Implementation effect: In urban road scenarios, scanning efficiency increased by 35%, the response time to sudden obstacles was ≤5ms, and the point cloud data integrity reached 99.2%.

[0120] Example 2

[0121] Multi-parameter collaborative dynamic energy efficiency management method

[0122] Implementation scenario: LiDAR for industrial inspection robots needs to balance scanning accuracy and power consumption during long-term operations.

[0123] Implementation steps

[0124] Offline training of power consumption models

[0125] Collect the scanning data of the robot under different working conditions (steering angular acceleration a=0.1~10rad / s2 , laser frequency f = 10 ~ 100kHz, ambient temperature T = 20 ~ 50 ℃), the power consumption prediction model is trained through the FPGA's built-in 3-layer neural network (input layer 3 nodes, hidden layer 8 nodes, output layer 1 node), and the weights are solidified into a multiplier-accumulator array.

[0126] Real-time energy efficiency index calculation

[0127] When the robot scans the storage shelves, the FPGA calculates the energy efficiency index E=0.8·a in real time. 2 / f+0.2·T (k1=0.8, k2=0.2) When E exceeds a threshold (preset value 15), the power regulation mechanism is triggered.

[0128] Hierarchical power regulation and gain compensation

[0129] First, the DAC controller reduces the laser power in 0.5W steps (from 5W to the 3W safety threshold), while simultaneously increasing the receiver gain by a factor of 1.2 (from 20dB to 24dB), maintaining a signal-to-noise ratio of 15dB or higher. A temperature sensor provides real-time feedback on the T value, dynamically adjusting the adjustment range.

[0130] Sleep policy activation

[0131] If no movement of goods in the shelf area is detected for five consecutive scanning cycles (each cycle is 10 seconds), it switches to low-power mode: the scanning path is reduced to the key area of ​​the shelf (depth 2m × width 1.5m), the laser power is reduced to 12% of the rated value (0.6W), and the MEMS actuator switches to stepping mode (scanning every 2 seconds).

[0132] Implementation effect: In an 8-hour inspection task, power consumption was reduced by 42%, and battery life was extended to 12 hours, while maintaining shelf contour detection accuracy ≤ 5mm.

[0133] Example 3

[0134] FPGA hierarchical scheduling and data compression transmission method

[0135] Implementation scenario: LiDAR for intelligent traffic intersections, which needs to process high-density point cloud data and achieve real-time scheduling.

[0136] Implementation steps

[0137] Point cloud data compression transmission

[0138] The original point cloud collected by the environment perception module (density is about 100 points / m 3After filtering through a voxel grid (resolution 10 cm × 10 cm × 10 cm), the data is compressed to 1 / 8 of its original size using 3D differential encoding (recording coordinate differences). The coprocessor extracts the outlines of obstacles (such as traffic light poles and zebra crossing edges) into a set of keypoint coordinates (≤10 points per obstacle) and writes them directly to the FPGA's dual-port RAM via a DMA channel.

[0139] Hierarchical scheduling modules work together

[0140] Path calculation layer: A 4-stage pipeline is used to parallel process grid partitions (16×16×16 grids are processed per cycle), the CORDIC algorithm is used to accelerate the calculation of spiral path parameters, and scanning instructions with a frequency of 100kHz are generated.

[0141] Power consumption decision layer: The lookup table stores 256 sets of (a, f, T) triples corresponding to the optimal power consumption configuration. When the steering angular acceleration a=5rad / s 2 , laser frequency f = 50kHz, T = 30℃, look up the table to find the configuration of output laser power 4W and receiving gain 22dB.

[0142] Hardware driver layer: A double-buffered mechanism alternately outputs scanning instructions (to the MEMS driver, with an update rate of 1MHz) and energy efficiency instructions (to the laser modulator, with an SPI rate of 10Mbps). The reconfigurable logic unit dynamically adjusts the laser pulse duty cycle (with an accuracy of 1ns).

[0143] Quick response to emergencies

[0144] When a vehicle is detected running a red light and entering an intersection, the sub-path replanning circuit generates an obstacle-avoiding path within 2ms, while compensating for the interrupted point cloud through an incremental coordinate counter to ensure a complete record of the traffic event.

[0145] Implementation effect: During peak traffic hours (traffic volume ≥ 200 vehicles / h), point cloud processing delay is ≤10ms, system power consumption fluctuation is ≤±5%, and the accuracy rate of abnormal event detection reaches 99.8%.

[0146] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. FPGA-based laser radar beam scanning path optimization and energy efficiency management method, characterized by: The following steps are involved: S1. Collect point cloud data and obstacle distribution information of the target area in real time through the environmental perception module; S2. Dynamic scanning path generation based on spatial partitioning: the target area is divided into priority grids within the FPGA. The historical scanning data and real-time obstacle positions are combined to calculate the scanning value weight of each grid cell and generate an adaptive spiral scanning path that avoids obstacles. S3, perform multi-parameter coordinated energy efficiency adjustment, dynamically matching laser transmission power, receiver gain and MEMS drive voltage according to the steering angular velocity and laser ranging frequency of the current scanning path; S4. Through the hierarchical scheduling module of FPGA, the optimized scanning path and energy efficiency instructions are synchronously output to the laser driving circuit and beam deflection controller.

2. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 1, characterized in that: The adaptive spiral scanning path generation in step S2 includes: establishing a spatial cost function for the priority grid C=α·D+β·I Where D is the Euclidean distance between the grid and the radar, I is the information entropy attenuation coefficient of the historical scan data, and α and β are weighting factors dynamically adjusted by the FPGA; Taking the minimum value of the cost function as the scanning starting point, a spiral path with a continuously variable radius is generated, and a real-time avoidance interpolation algorithm for sudden obstacles is embedded.

3. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 1, characterized in that: The dynamic energy efficiency adjustment in step S3 includes: The energy efficiency index of the path steering angular acceleration a and the laser emission frequency f is calculated using the power consumption model built into the FPGA. E=k1·a 2 / f+k2·T Where k1 and k2 are calibration coefficients, and T is the real-time data of the temperature sensor; When E exceeds the preset threshold, the laser power is gradually reduced to a safe threshold, and the receiver gain is increased proportionally to maintain the signal-to-noise ratio.

4. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 1, characterized in that: The FPGA hierarchical scheduling module includes: Path calculation layer: uses a parallel pipeline structure to perform space partitioning and path generation; Power consumption decision layer: An integrated lookup table stores the optimal power consumption configuration for different steering angle-laser frequency combinations; Hardware driver layer: Scanning instructions and energy efficiency instructions are alternately output to the laser and MEMS driver through a double buffer mechanism.

5. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 2, characterized in that: The sudden obstacle avoidance algorithm is: A monitoring window is embedded in the scanning path. When a new obstacle is detected in the real-time point cloud, the current scanning task is terminated. The sub-path replanning circuit is activated in the FPGA to generate a local B-spline curve alternative path with the obstacle contour as the constraint boundary and compensate for the interrupted point cloud data.

6. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 3, characterized in that: The power consumption model is generated through offline training: Collect current sensor data under different scanning modes and build a power consumption prediction neural network in the FPGA with angular acceleration, laser frequency, and ambient temperature as input; Solidify the neural network weights as a multiplier-accumulator array in hardware logic circuits.

7. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 1, characterized in that: Also includes: Start the energy-efficient sleep strategy during the lidar idle phase: When no moving target is detected for N consecutive scanning cycles, it switches to low-power monitoring mode: The scanning path is reduced to the preset key area, the laser emission power is reduced to the sleep threshold, and the MEMS actuator is switched to stepping mode operation.

8. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 4, characterized in that: The hardware driver layer contains reconfigurable logic units: Dynamically configure the duty cycle accuracy of laser pulse encoding to 1ns level; The 12-bit high-precision PWM module generates an adaptive MEMS drive waveform to reduce steering mechanical losses.

9. The FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to claim 1, characterized in that: Data compression is used for transmission between the environment perception module and the FPGA: The point cloud data is filtered and differentially encoded by the voxel grid and then transmitted to the FPGA; The obstacle outline is extracted by the edge detection coprocessor as a set of key point coordinates and written directly to the FPGA memory through the DMA channel.

10. A laser radar system, applied to the FPGA-based laser radar beam scanning path optimization and energy efficiency management method according to any one of claims 1 to 9, characterized in that: include: FPGA chip, multi-line laser emission array, avalanche photodiode receiving array, two-dimensional MEMS micromirror, and temperature-current sensor group; The FPGA chip integrates a scan path optimization pipeline, an energy efficiency decision state machine, and a multi-channel PWM hardware controller.

Images