SLAM system with adaptive energy consumption management and energy consumption optimization method

Abstract

The invention discloses an SLAM system with adaptive energy consumption management and an energy consumption optimization method, and the system comprises an energy consumption sensing module, a task demand analysis module, an environment complexity evaluation module and an adaptive adjustment module. The calculation complexity of the SLAM algorithm and the use frequency of the sensor are dynamically adjusted, the energy consumption management strategy of the equipment is optimized by combining the task priority and environment perception, the endurance time of the equipment is prolonged, calculation and sensor resource use are adaptively adjusted on the premise of ensuring the performance of the SLAM system, and the service life of the equipment is prolonged. And the balance between the optimized SLAM performance and the energy consumption is realized in different scenes.

Description

Technical Field

[0001] The present invention belongs to the field of navigation technology, relates to a synchronous positioning and mapping technology, and specifically relates to a SLAM system with adaptive energy consumption management and an energy consumption optimization method. Background Art

[0002] With the widespread use of devices such as mobile robots, smartphones and drones, SLAM technology plays a key role in autonomous navigation, environmental perception and augmented reality. However, these devices are usually limited by battery capacity, which may significantly shorten the device's battery life when performing complex computing tasks. Traditional SLAM algorithms usually run at a fixed computational complexity and sensor sampling frequency, and fail to consider the device's real-time power status, task priority and environmental complexity, resulting in inefficient energy consumption. Summary of the invention

[0003] Technical purpose: In response to the above technical problems, the present invention proposes a SLAM system with adaptive energy management and an energy optimization method, which can dynamically adjust the computational complexity of the SLAM algorithm and the frequency of sensor use according to the real-time power status of the device, task requirements and environmental complexity, and optimize the energy management strategy of the device by combining task priority and environmental perception. Under the premise of ensuring the performance of the SLAM system, the battery life of the device can be significantly extended. It can adaptively adjust the use of computing and sensor resources and optimize the SLAM algorithm framework for energy management.

[0004] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution:

[0005] A SLAM system with adaptive energy management, comprising:

[0006] Energy consumption sensing module, used to obtain battery parameters, including battery voltage, current and temperature, and calculate energy consumption sensing results in real time, including remaining power and estimated battery life;

[0007] The task requirement analysis module is used to load the current task information, including task type, priority, performance requirements and expected completion time, and convert the task requirements into quantitative SLAM performance indicators;

[0008] An environmental complexity assessment module is used to obtain environmental information and output an assessment result of environmental complexity;

[0009] The adaptive adjustment module is used to receive information sent by the energy consumption perception module, the task requirement analysis module and the environment complexity assessment module, and formulate an energy consumption optimization strategy based on the received information.

[0010] Preferably, in the task requirement analysis module, tasks are classified based on preset rules, and parameter selection and behavior decisions under different task types are defined. Parameter selection includes selecting quantified navigation accuracy.

[0011] Preferably, the environment complexity assessment module includes:

[0012] A data acquisition unit, including a camera for acquiring images and a laser radar for acquiring point cloud data;

[0013] A feature extraction unit, including an image feature extraction unit, for extracting environmental image features from the collected image, including edges, corners and texture complexity; and a point cloud feature extraction unit, for calculating environmental point cloud features, including point cloud density, plane and surface distribution, based on the collected point cloud data;

[0014] A point cloud feature extraction unit, used to calculate the environmental point cloud features, including point cloud density, plane and surface distribution, based on the collected point cloud data;

[0015] The complexity evaluation unit is used to calculate the environment complexity score or level according to the environment image features and the environment point cloud features.

[0016] Preferably, the adaptive adjustment module includes:

[0017] The parameter adjustment strategy formulation unit is used to determine the adjustment direction, including the adjustment of calculation complexity and the adjustment of sensor usage frequency, based on the energy consumption perception results, task requirements and environmental complexity evaluation results, and with reference to the preset strategy table;

[0018] The algorithm start-stop unit is used to control the frequency of loop detection. When the environment complexity is lower than the preset value, it simplifies or skips the complex feature extraction, or when the task allows, it reduces the frequency of loop detection to reduce the amount of calculation.

[0019] Preferably, the computational complexity adjustment includes particle filter SLAM adjustment, optimization iteration number adjustment and map resolution adjustment, wherein the particle filter SLAM adjustment is to control the amount of calculation by adjusting the number of particles, increasing or decreasing the number of particles, the optimization iteration number adjustment is to dynamically adjust the number of iterations or termination conditions in the optimization algorithm, and the map resolution adjustment is used to adjust the grid size of the map as needed;

[0020] The sensor usage frequency adjustment includes sampling rate control adjustment and sensor selective usage adjustment, wherein the sampling rate control adjustment is to modify the sampling frequency of the camera and lidar, and the sensor selective usage adjustment is to turn off sensors with high power consumption when the environment is simple or the battery is low.

[0021] Preferably, the energy consumption optimization strategy includes a performance priority mode, a balance mode and an energy-saving mode;

[0022] The triggering conditions of the performance priority mode are: power level above 70%, high task priority, high environment complexity, and the adjustment measures are: maximizing computational complexity, using high-precision algorithms, and increasing sensor sampling frequency;

[0023] The triggering conditions of the balance mode are: power level between 40% and 70%, medium task priority, medium environment complexity, and the adjustment measures are: reducing the computational complexity and positioning accuracy to a preset range, and adjusting to a medium sensor sampling frequency;

[0024] The triggering conditions of the energy-saving mode are: power level below 40%, low task priority, and low environmental complexity. The adjustment measures are: reducing the calculation complexity to a preset lower limit, using a simplified algorithm, reducing the sensor sampling frequency, or turning off the sensor at a preset position.

[0025] A method for optimizing energy consumption of a SLAM system with adaptive energy consumption management is used in the system, and the method comprises the following steps:

[0026] Real-time monitoring: continuously run the energy consumption perception module, task requirement analysis module and environment complexity assessment module to obtain real-time data;

[0027] Strategy selection: The adaptive adjustment module selects the most appropriate energy consumption optimization strategy from the preset strategy library based on real-time data;

[0028] Parameter adjustment: Execute the adjustment measures in the strategy to dynamically modify the SLAM algorithm parameters and sensor working status.

[0029] Preferably, the method further comprises the steps of:

[0030] Feedback and iteration: The system monitors the effects of adjustments and automatically adjusts the energy consumption optimization strategy if any abnormal situation occurs.

[0031] Beneficial effects: Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0032] (1) The present invention innovatively combines real-time energy consumption perception with the SLAM algorithm, enabling the system to dynamically adjust the algorithm behavior according to the power status. The energy consumption perception module not only monitors the current power but also predicts the power consumption trend, providing a more accurate reference for subsequent strategy adjustments.

[0033] (2) The present invention proposes an adaptive adjustment strategy guided by task requirements and environmental complexity, directly mapping task requirements into performance parameters of the SLAM algorithm, enabling the system to be optimized for different tasks and using machine learning algorithms (such as support vector machines or neural networks) to classify environmental complexity, which is more accurate and robust than traditional methods and can intelligently optimize the balance between SLAM performance and energy consumption in different scenarios.

[0034] (3) The present invention provides a multi-level energy consumption optimization method by adjusting the computational complexity, the frequency of sensor use and the activation of algorithm modules, further improving the flexibility and adaptability of the system. Based on multi-dimensional adaptive adjustment, it can control various aspects of the SLAM algorithm in a fine-grained manner, achieving an energy consumption management effect that traditional SLAM systems cannot achieve. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The present invention proposes a structural schematic diagram of a SLAM system with adaptive energy consumption management;

[0036] Figure 2 for Figure 1 Schematic diagram of the system applied to the robot;

[0037] Figure 3 for Figure 1 Flowchart of the energy consumption optimization method implemented by the system. DETAILED DESCRIPTION

[0038] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] like Figure 1 As shown, the present invention proposes a SLAM system with adaptive energy consumption management, comprising the following modules:

[0040] Energy consumption sensing module, used to obtain battery parameters, including battery voltage, current and temperature, and calculate the remaining power and estimated battery life in real time;

[0041] The task requirement analysis module is used to load the current task information, including task type, priority, performance requirements and expected completion time, and convert the task requirements into quantitative SLAM performance indicators;

[0042] Environmental complexity assessment module, used to obtain environmental information;

[0043] The adaptive adjustment module is used to receive information sent by the energy consumption perception module, the task requirement analysis module and the environment complexity assessment module, and formulate an energy consumption optimization strategy based on the received information.

[0044] like Figure 2As shown, the system is mounted on a robot. The robot is equipped with a power monitoring sensor for monitoring its own battery power, a SLAM algorithm unit for realizing positioning and mapping functions, a controller / microprocessor with a task management system, and a camera and lidar for detecting the robot's environmental information.

[0045] Combination Figure 3 The detailed workflow is as follows:

[0046] 1. System initialization:

[0047] After the robot starts, the SLAM system loads the following modules:

[0048] Energy consumption sensing module: Initializes the power monitoring sensor and obtains initial power data.

[0049] Task requirement analysis module: loads current task information, including task type, priority and expected completion time.

[0050] Environmental complexity assessment module: prepares the sensor data processing pipeline and initializes the environmental feature extraction algorithm.

[0051] Adaptive adjustment module: Set default parameters and prepare for dynamic adjustment based on subsequent data.

[0052] 2. Each module starts working and monitors in real time

[0053] The system continuously runs the energy consumption perception module, task requirement analysis module and environmental complexity assessment module to obtain real-time data.

[0054] 2.1. Energy consumption perception module workflow:

[0055] Real-time monitoring unit: reads battery parameters such as voltage, current and temperature for real-time calculation of remaining power and estimated battery life.

[0056] Energy consumption prediction unit: Utilizes historical power consumption data and current load conditions, and applies prediction algorithm models (such as linear regression models or support vector machines SVM, etc.) to predict the power consumption rate in the future.

[0057] 2.2. Task Requirements Analysis Module Workflow:

[0058] Task parsing unit: obtains task details from the task management system, for example:

[0059] Mission type: emergency material delivery, regional inspection, environmental monitoring, etc.

[0060] Priority: high, medium, low;

[0061] Performance requirements: positioning accuracy, map update frequency, obstacle avoidance requirements, etc.

[0062] Demand quantification unit: Converts task requirements into quantitative SLAM performance indicators, such as the minimum positioning accuracy that needs to be achieved (such as ±10 cm).

[0063] Navigation accuracy is the parameter that has the greatest impact on SLAM performance. Higher navigation accuracy requires high-frequency collection of sensor data, positioning calculations, and detailed attitude control, all of which will greatly affect energy consumption. According to actual tests, the navigation accuracy of ±1cm is compared to the navigation accuracy of ±3cm. Under the same resource occupation, the time consumption, including positioning calculations and attitude control at the point, will increase by more than 200%. The navigation accuracy of ±5cm will increase the time consumption by more than 100% compared to the navigation accuracy of ±3cm.

[0064] The strategies that have a greater impact include planning and positioning. Planning can be divided into dynamic autonomous path planning and static preset line planning. Positioning modes can be divided into wheel speed + IMU, wheel speed + LIO / VIO, wheel speed + LIVO, wheel speed + IMU / LIO / VIO / LIVO + positioning map registration, wheel speed + IMU + RTK and other combinations. More complex planning or positioning strategies require more sensors to work, which will increase computing consumption while improving the robustness of the navigation system.

[0065] Therefore, the present invention analyzes the task requirements and dynamically adjusts various parameters affecting navigation accuracy, including map resolution, positioning calculation step / frequency, number of positioning calculation iterations, sensor data acquisition frequency, and in-place posture threshold, according to the task requirements, so as to effectively optimize the performance and energy consumption of the SLAM system.

[0066] Task requirements are classified based on preset rules and define parameter selection and behavior decisions for different types. For example, tasks such as fixed-point mark recognition and defect detection require high-precision navigation accuracy of ±1cm. Infrared temperature measurement tasks can relax the navigation accuracy to ±3cm. Environmental monitoring, audio analysis and other tasks have lower navigation accuracy requirements, which can be relaxed to ±5cm. Mobile inspection tasks are different from fixed-point inspection tasks. In addition to the requirements for navigation accuracy, they also have requirements for real-time performance. Usually, the posture recursive model will be based on the multi-sensor fusion model of wheel speed + IMU instead of wheel speed + IMU + LO + VO.

[0067] 2.3. Workflow of the environment complexity assessment module:

[0068] Data acquisition unit: collects camera images and lidar point cloud data.

[0069] Feature extraction unit: uses computer vision and point cloud processing algorithms to extract environmental features, including:

[0070] Image features: extract information such as edges, corners, texture complexity, etc.;

[0071] Point cloud features: Calculate point cloud density, plane and surface distribution and other indicators.

[0072] Complexity assessment unit: Based on the above characteristics, the environment complexity score is calculated.

[0073] The determination of environmental complexity is relatively complex. Except for some simple corridor environments and open environments that can be distinguished by traditional methods, more scene complexities cannot be simply classified based on predetermined rules. The present invention designs a method based on machine learning, including the following steps:

[0074] (1) Define the environmental complexity index

[0075] In machine learning-based evaluation, we first need to clarify how to define environmental complexity and convert it into learnable features or labels. For example, visual complexity, geometric complexity, and dynamic complexity. Visual complexity includes texture richness or sparseness, lighting conditions (lighting changes, shadows, etc.), and the density and distribution of feature points. Geometric complexity includes the complexity of spatial structure (such as the density of obstacles and geometric diversity) and the openness or narrowness of the scene. Dynamic complexity includes the number and distribution of dynamic objects and the flow of people or vehicles in the environment.

[0076] (2) Data collection and annotation

[0077] First, collect data and annotate the complexity of the environment for training machine learning models. The data source can usually be public data sets on the Internet (KITTI, TUM RGB-D, NYUv2, etc.) and sensor data collected on site (lidar point cloud, visual camera image frames, IMU data, odometer data, etc.). In addition, a large amount of synthetic data can be generated through simulated environments (such as Gazebo, CARLA). The annotation method is to assign complexity labels (such as "simple", "medium complexity", "complex") to each environment through manual annotation in the early stage. When a certain amount of initial data is available, unsupervised learning or heuristic rules can be used to automatically generate complexity labels (such as automatically estimating complexity based on the number of feature points or occlusion rate).

[0078] (3) Feature extraction

[0079] In order for machine learning models to understand the complexity of the environment, effective features need to be extracted. In addition to counting the number, distribution, and quality of feature points based on traditional feature point detectors such as SIFT and ORB, image features can also be based on convolutional neural networks (CNN) to extract global or local features of images. In addition to geometric features such as point cloud density, plane characteristics, and curvature distribution, point cloud features can also be directly processed using deep learning models such as PointNet. Temporal features usually use recurrent neural networks (RNN) or Transformer to process the dynamic changes of the environment.

[0080] (4) Model training and evaluation

[0081] Based on the features and labeled data, the machine learning model is trained to predict the complexity of the environment. Machine learning can generally be divided into supervised learning, unsupervised learning, and reinforcement learning. Supervised learning uses classification models (such as random forests, support vector machines, and deep neural networks) to predict the category of environmental complexity. Use regression models to predict continuous complexity scores. Unsupervised learning analyzes the distribution of environmental features and divides the complexity through clustering (such as K-Means and DBSCAN). Use models such as autoencoders to learn low-dimensional representations of environmental features and evaluate complexity based on feature distribution. Reinforcement learning incorporates complexity evaluation into SLAM strategies and learns the impact of complexity by interacting with the environment.

[0082] (5) Combined with SLAM performance indicators

[0083] Correlate the environmental complexity with the SLAM performance indicators to further verify the effectiveness of the evaluation. For example:

[0084] Positioning Error: Complex environments may lead to higher positioning errors.

[0085] Map quality: Complex environments may cause map discontinuities or distortions.

[0086] Data loss rate: Complex environments may cause sensor data loss (such as occlusion in vision).

[0087] Computational resource consumption: Complex environments may increase the computational overhead of SLAM algorithms.

[0088] By analyzing these performance indicators, machine learning models can further optimize the accuracy of complexity assessment.

[0089] 2.4. Specific operations of the adaptive adjustment module:

[0090] Parameter adjustment strategy formulation unit: Based on the energy consumption perception results, task requirements and environmental complexity, refer to the preset strategy table to determine the adjustment direction, including:

[0091] Computational complexity adjustment:

[0092] Particle filter SLAM: adjust the number of particles, increase or decrease the number of particles to control the amount of calculation;

[0093] Optimize the number of iterations: In the optimization algorithm, dynamically adjust the number of iterations or termination conditions;

[0094] Map Resolution: Adjust the map's grid size as needed.

[0095] Sensor usage frequency adjustment:

[0096] Sampling rate control: modify the sampling frequency of the camera and lidar, such as reducing the sampling rate of the lidar from 10Hz to 5Hz;

[0097] Selective use of sensors: When the environment is simple or the battery is low, turn off sensors that consume a lot of power, such as lidar, and only use the camera;

[0098] Loop detection module: performs loop detection;

[0099] Algorithm module start and stop unit:

[0100] Feature extraction module: In low-complexity environments, simplify or skip complex feature extraction. If the task allows, reduce the frequency of loop detection to reduce the amount of computation.

[0101] Loop detection module: When the task allows, reduce the frequency of loop detection to reduce the amount of computation.

[0102] 3. Strategy selection

[0103] The adaptive adjustment module selects the most appropriate energy consumption optimization strategy from the preset strategy library according to real-time data, executes the adjustment measures in the strategy, and dynamically modifies the SLAM algorithm parameters and sensor working status.

[0104] The energy consumption optimization strategy examples are as follows:

[0105] Strategy (1): Performance-first mode

[0106] Trigger conditions: battery level above 70%, high task priority, and high environment complexity.

[0107] Adjustment measures:

[0108] Maximize computational complexity and use high-precision algorithms.

[0109] Increase sensor sampling frequency to ensure rich data.

[0110] Expected results: Ensure the highest SLAM performance and meet stringent mission requirements.

[0111] Strategy (2): Balanced mode

[0112] Trigger conditions: battery level between 40% and 70%, medium task priority, and medium environment complexity.

[0113] Adjustment measures:

[0114] Appropriately reduce the computational complexity and maintain reasonable positioning accuracy.

[0115] Medium sensor sampling frequency, balancing data quality and energy consumption.

[0116] Expected effect: Extend the battery life while ensuring the completion of the task.

[0117] Strategy (3): Energy-saving mode

[0118] Trigger conditions: battery power is less than 40%, task priority is low, and environment complexity is low.

[0119] Adjustment measures:

[0120] Greatly reduce the computational complexity and adopt a simplified algorithm.

[0121] Reduce the sensor sampling frequency or even turn off some sensors.

[0122] Expected results:

[0123] Save energy as much as possible and extend equipment running time.

[0124] 4. Parameter adjustment

[0125] Execute the adjustment measures in the strategy to dynamically modify the SLAM algorithm parameters and sensor working status.

[0126] 5. Feedback and iteration

[0127] The system monitors the effect of the adjustment. If any abnormality occurs (such as low positioning accuracy), the strategy will be automatically adjusted to ensure system stability.

[0128] The present invention integrates task requirements and environmental complexity to obtain an effective SLAM system that balances performance and energy consumption. Based on preset parameters based on task requirements, the complexity assessment based on machine learning is used as an online module to dynamically adjust the parameters and strategies of the SLAM algorithm, including dynamically adjusting the feature extraction method: switching to a more robust feature extraction algorithm in a complex environment. Optimizing sensor selection and fusion: dynamically adjusting sensor weights (such as the fusion of vision and lidar) according to environmental complexity. Path planning and repositioning: selecting a wider path or adding a repositioning module in a complex environment. Increasing the acquisition and operation frequency of perception sensors in a highly dynamic environment. More accurate and robust than traditional methods, it can intelligently optimize the balance between SLAM performance and energy consumption in different scenarios.

[0129] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A SLAM system with adaptive energy management, characterized in that: include: Energy consumption sensing module, used to obtain battery parameters, including battery voltage, current and temperature, and calculate energy consumption sensing results in real time, including remaining power and estimated battery life; The task requirement analysis module is used to load the current task information, including task type, priority, performance requirements and expected completion time, analyze the task requirements and convert them into quantitative SLAM performance indicators; An environmental complexity assessment module is used to obtain environmental information and output an assessment result of environmental complexity; The adaptive adjustment module is used to receive information sent by the energy consumption perception module, the task requirement analysis module and the environment complexity assessment module, and formulate an energy consumption optimization strategy based on the received information.

2. A SLAM system with adaptive energy management according to claim 1, characterized in that: In the task requirement analysis module, tasks are classified based on preset rules, and parameter selection and behavior decisions under different task types are defined. Parameter selection includes selecting quantified navigation accuracy.

3. A SLAM system with adaptive energy management according to claim 1, characterized in that: The environment complexity assessment module includes: A data acquisition unit, including a camera for acquiring images and a laser radar for acquiring point cloud data; A feature extraction unit, including an image feature extraction unit, for extracting environmental image features from the collected image, including edges, corners and texture complexity; and a point cloud feature extraction unit, for calculating environmental point cloud features, including point cloud density, plane and surface distribution, based on the collected point cloud data; The complexity evaluation unit is used to calculate the environment complexity score or level according to the environment image features and the environment point cloud features.

4. A SLAM system with adaptive energy management according to claim 1, characterized in that: The adaptive adjustment module comprises: The parameter adjustment strategy formulation unit is used to determine the adjustment direction, including the adjustment of calculation complexity and the adjustment of sensor usage frequency, based on the energy consumption perception results, task requirements and environmental complexity evaluation results, and with reference to the preset strategy table; The algorithm start-stop unit is used to control the frequency of loop detection. When the environment complexity is lower than the preset value, it simplifies or skips the complex feature extraction, or when the task allows, it reduces the frequency of loop detection to reduce the amount of calculation.

5. A SLAM system with adaptive energy management according to claim 4, characterized in that: The computational complexity adjustment includes particle filter SLAM adjustment, optimization iteration number adjustment and map resolution adjustment, wherein the particle filter SLAM adjustment is to increase or decrease the number of particles by adjusting the number of particles to control the amount of calculation, the optimization iteration number adjustment is to dynamically adjust the iteration number or termination condition in the optimization algorithm, and the map resolution adjustment is used to adjust the grid size of the map as needed; The sensor usage frequency adjustment includes sampling rate control adjustment and sensor selective usage adjustment, wherein the sampling rate control adjustment is to modify the sampling frequency of the camera and lidar, and the sensor selective usage adjustment is to turn off sensors with high power consumption when the environment is simple or the battery is low.

6. A SLAM system with adaptive energy management according to claim 1, characterized in that: The energy consumption optimization strategy includes performance priority mode, balance mode and energy saving mode; The triggering conditions of the performance priority mode are: power level above 70%, high task priority, high environment complexity, and the adjustment measures are: maximizing computational complexity, using high-precision algorithms, and increasing sensor sampling frequency. The triggering conditions of the balance mode are: power level between 40% and 70%, medium task priority, medium environment complexity, and the adjustment measures are: reducing the computational complexity and positioning accuracy to a preset range, and adjusting to a medium sensor sampling frequency; The triggering conditions of the energy-saving mode are: power level below 40%, low task priority, and low environmental complexity. The adjustment measures are: reducing the calculation complexity to a preset lower limit, using a simplified algorithm, reducing the sensor sampling frequency, or turning off the sensor at a preset position.

7. A method for optimizing energy consumption of a SLAM system with adaptive energy consumption management, used in any system of claims 1-6, characterized in that: The method comprises the following steps: Real-time monitoring: continuously run the energy consumption perception module, task requirement analysis module and environment complexity assessment module to obtain real-time data; Strategy selection: The adaptive adjustment module selects the most appropriate energy consumption optimization strategy from the preset strategy library based on real-time data; Parameter adjustment: Execute the adjustment measures in the strategy to dynamically modify the SLAM algorithm parameters and sensor working status.

8. The energy consumption optimization method of a SLAM system with adaptive energy consumption management according to claim 7, characterized in that: The method further comprises the steps of: Feedback and iteration: The system monitors the effects of adjustments and automatically adjusts the energy consumption optimization strategy if any abnormal situation occurs.

Images