Real-time motion planning system based on dynamic environment perception

Abstract

The invention provides a real-time motion planning system based on dynamic environment perception, and aims to solve the limitation of the prior art in a dynamic environment. According to the system, environment data are captured and processed in real time through a highly integrated sensor fusion technology, and an accurate environment model is generated. And in combination with a deep learning algorithm, the system can predict environment change and realize quick response. In addition, the system adopts an advanced dynamic path adjustment algorithm to ensure path optimization and safe navigation of the machine in a complex environment. Particularly, a multi-machine collaborative planning mechanism is introduced, and efficient collaboration and avoidance among machine groups are realized through a distributed consistency algorithm and multi-agent reinforcement learning. On the whole, the system obviously improves the application potential of the machine in a variable environment, enhances the adaptability and stability of the machine, and opens up a new possibility for the development of the intelligent machine equipment technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent machines, and in particular relates to a real-time motion planning system capable of perceiving a dynamic environment. Background Art

[0002] In the field of intelligent machine technology, motion planning is a key technology that ensures machines can move safely and efficiently from one point to another. Traditional motion planning methods, such as the A* algorithm, Dijkstra's algorithm, artificial potential field method, and RRT (Rapidly Exploring Random Trees) algorithm, while performing well in static environments, often struggle with dynamic ones. Factors such as moving obstacles and changing environmental structures in dynamic environments require machines to be able to perceive environmental changes in real time and quickly adjust their motion paths to avoid collisions and improve task execution efficiency.

[0003] While existing real-time motion planning algorithms, such as DWA (Dynamic Window Planning) and RRT*, take dynamic obstacles into account to some extent, they still face challenges in computational efficiency and planning accuracy when dealing with highly complex and dynamic environments. Furthermore, collaborative planning for multi-machine systems requires ensuring effective collaboration and optimal resource allocation while ensuring the performance of individual machines, a challenge that has yet to be fully addressed in existing technologies.

[0004] With the development of artificial intelligence technologies such as deep learning, data-driven approaches have brought new opportunities for machine motion planning. However, effectively integrating these advanced technologies into motion planning systems to achieve rapid response and adaptation to dynamic environments remains a hot topic and a challenge in this field. It is against this backdrop that this present invention aims to address the limitations of existing motion planning systems in dynamic environments through innovative technical means, thereby enhancing the autonomy and intelligence of machines.

[0005] Technical Solution

[0006] This invention provides a real-time motion planning system based on dynamic environment perception. This system adopts a series of innovative technical solutions to achieve efficient and safe motion planning in response to the complexity of dynamic environments. The following are the specific technical solutions of this system:

[0007] The multi-source sensor integration module is a key technological component for achieving precise environmental perception. This module is meticulously designed, integrating multiple sensor types, each optimized for specific environmental perception needs. The core of the module is the LiDAR sensor, which accurately measures the distance between the machine and surrounding objects by emitting and receiving laser pulses. This sensor generates high-resolution point cloud data, providing the machine with a detailed spatial map, which is essential for precise positioning and navigation. Stereo cameras provide the system with depth perception. By simulating human binocular vision, stereo cameras can measure the three-dimensional shape and distance of objects, identify objects in the environment, and assess their size. This enables the machine to more accurately understand the layout of objects in its operating environment. Ultrasonic sensors play a vital role in the machine's close-range perception. They detect and measure the distance between the machine and obstacles by emitting ultrasonic waves and measuring the duration of the echo. These sensors are particularly important in obstacle avoidance and precise docking. Infrared sensors can detect temperature changes and heat sources in the environment, providing the machine with a sensing capability that works in varying lighting conditions. Infrared sensors are particularly useful for nighttime operation or in environments with limited light. Force sensors, installed on the contact surfaces of machines, measure the contact force between the machine and its environment. These sensors provide tactile feedback, enabling more refined and natural interaction with the environment. By integrating data from these sensors, multi-source sensor integration modules can provide intelligent machines with comprehensive, multi-dimensional environmental perception. This comprehensive perception capability is a prerequisite for advanced motion planning and safe navigation, ensuring that machines can operate effectively in complex and dynamic environments.

[0008] Preferably, the raw data collected by sensors often contains noise, which may originate from the limitations of the sensors themselves or interference from the external environment. To improve data clarity and reliability, the advanced data preprocessing module first uses filtering algorithms to reduce noise. For example, the Kalman filter can effectively remove random noise from the data based on prediction and observation models, retaining valuable signals. Since data from different sensors may have different dimensions and scales, directly using this data can lead to difficulties in analysis and processing. To this end, the advanced data preprocessing module normalizes the data, converting it to a unified numerical range. This step not only eliminates dimensional differences but also enhances data comparability, laying the foundation for subsequent data processing and analysis. Data synchronization is a critical issue in multi-sensor systems. Since sensors may collect data at slightly different times, these discrepancies can lead to misunderstandings about environmental conditions. The advanced data preprocessing module ensures temporal consistency of data from different sensors through timestamp calibration or event-triggered mechanisms. This allows the machine to accurately understand the environmental conditions at a specific moment. The advanced data preprocessing module uses data fusion technology to integrate information from different sensors to form a comprehensive and consistent environmental perception model. Data fusion can be performed at multiple levels, including sensor-level fusion (directly merging raw data), feature-level fusion (merging processed features), and decision-level fusion (making a final decision based on the output of different sensors). Through data fusion, machines can obtain richer and more robust environmental information and improve their perception of complex environments. Through this series of processing steps, the advanced data preprocessing module transforms raw, potentially noisy data from different sensors into accurate, consistent, and usable information, providing a solid data foundation for intelligent machine motion planning systems.

[0009] Preferably, the Environment Perception and Modeling module first fuses data from various sensors. This step involves combining LiDAR point cloud data, stereo camera depth images, and possibly other sensor data to form a unified representation of the environment. The data fusion algorithm employs optimization techniques to ensure consistency and complementarity between the various data sources. Leveraging computer vision technology, the module is able to identify and classify objects in the environment. Using deep learning algorithms, such as convolutional neural networks (CNNs), machines can extract features from visual data to identify and classify elements such as obstacles, ground surfaces, and walls. The module utilizes SLAM technology, a technique that allows machines to localize and map objects in unknown environments. SLAM algorithms combine sensor data with machine motion information to update the machine's position and environmental map in real time. For example, visual SLAM or laser SLAM technology is used, depending on the sensor type. Combining data fusion, computer vision, and SLAM technologies, the Environment Perception and Modeling module generates a highly accurate three-dimensional model. This model not only includes the geometry of the environment but may also include semantic information about objects, such as obstacle types and travel zones. The 3D model can achieve centimeter-level accuracy, providing the necessary spatial information for machine motion planning. When using visual SLAM technology, camera parameters such as focal length, resolution, and frame rate are precisely set to ensure image data quality. At the same time, initialization parameters in the SLAM algorithm, such as feature extraction thresholds, map resolution, and keyframe selection strategies, are carefully tuned to suit different environmental conditions.

[0010] Preferably, the dynamic obstacle detection and classification module uses a lightweight CNN model with a processing speed of at least 30 frames per second to ensure synchronization with the machine's movement. The model training uses the cross-cutting loss function To optimize category distinction, the formula is expressed as: Among them, C represents the total number of categories, y o,c is the one-hot encoding of the true label, y p,c is the probability predicted by the model. The model uses the Adam optimizer with a learning rate of 0.001 and a batch size of 32 to achieve fast convergence and high generalization. After training, the model achieved over 95% accuracy on the test set and completed the inference process for a single frame in 200 milliseconds. Furthermore, for tracking dynamic obstacles, the module uses deep learning-based tracking algorithms such as YOLO or SSD, which are capable of handling fast-moving obstacles and provide information on their trajectory and velocity.

[0011] Preferably, the prediction analysis module uses recurrent neural networks (RNN) and reinforcement learning technology to accurately predict the future movement trajectory of obstacles and environmental change trends. The module first processes the time series data through RNN to capture the historical movement pattern of obstacles and uses the following formula to update the state at each time t in the sequence: t =f(h t-1 ,x t ) where h t represents the hidden state at time t, x t is the input of the current time step, and f is the recursive function of the RNN, usually a tanh or ReLU activation function. Then, based on the learned time dependency, the module uses a reinforcement learning algorithm, such as the Deep Q Network (DQN), to predict the optimal motion strategy for the obstacle. Its objective function can be expressed as: Here, π represents the strategy, r t is the reward at time t, and γ is a discount factor that balances the importance of immediate rewards and future rewards. Through this approach, the predictive analysis module can predict the likely locations of obstacles in the next few time steps, providing forward-looking information for the machine's path planning and enhancing its adaptability and safety in dynamic environments.

[0012] The real-time path optimization module preferably uses a modified A* algorithm or D*Lite algorithm as the core of its path search. The A* algorithm is an efficient path search algorithm that quickly finds the optimal path by evaluating the cost from the starting point to the destination and combining it with a heuristic function to estimate the cost of reaching the destination. The D*Lite algorithm is a path replanning algorithm suitable for dynamic environments, capable of rapidly adjusting the existing path when the environment changes, rather than recalculating the entire path. The real-time path optimization module is tightly integrated with the environmental perception and modeling module, using the high-precision three-dimensional model generated by the module to construct the search space. This model provides precise information on the location and shape of obstacles, ensuring the accuracy of the path search. The module incorporates the output of the deep learning-based predictive analysis module to take into account the future location of obstacles. Path optimization is based not only on the current environmental state but also on dynamic environmental changes, improving the adaptability and foresight of the path. To meet real-time requirements, the module has implemented multiple performance optimizations. These include the use of efficient data structures, such as priority queues, to quickly access and update path nodes, and parallel computing techniques that leverage the multi-core capabilities of modern processors to accelerate the path search process. The A* or D*Lite algorithm in the module can be parameterized based on the machine's operating conditions and environmental characteristics. The search depth and frequency can be adjusted based on the machine's movement speed and environmental complexity to ensure the optimal path is obtained in any situation.

[0013] Preferably, the kinematic and dynamic constraint processing module analyzes parameters such as the machine's joint angle limits, maximum speed, and acceleration. For example, it sets the rotation range of each joint of the machine to between -180° and 180°, the maximum linear speed to 1.5 m / s, and the maximum acceleration to 0.5 m / s2. Through strict verification of these kinematic parameters, the module automatically adjusts the path points to avoid exceeding the machine's motion capabilities. At the same time, it uses a dynamic model to evaluate the energy consumption and stability of the path to ensure the safety and efficiency of the machine when executing the path. In addition, the module also integrates a collision detection algorithm to monitor potential collision risks on the path in real time and make necessary adjustments to the obstacle avoidance strategy to achieve smooth, coherent, and physically compliant motion planning.

[0014] Preferably, the emergency obstacle avoidance and safety assurance module integrates advanced collision detection algorithms, such as collision detection based on geometric models, which can monitor the distance between the machine and obstacles in real time. The safety distance threshold is set to 0.5 meters. Once the distance between the machine and the obstacle is detected to be less than the threshold, the system triggers an alarm. At the same time, the module adopts a safety distance assessment mechanism to evaluate the current speed of the machine and the movement state of the obstacle, and calculate the minimum safety distance. When a potential collision risk is detected, the module can quickly generate obstacle avoidance strategies, such as slowing down, stopping or changing the direction of movement, and by executing these strategies, ensure the safety of the machine and the surrounding environment. In addition, the module also has self-learning capabilities, which can continuously optimize the obstacle avoidance strategy based on historical obstacle avoidance data to improve the safety and reliability of the machine in complex environments.

[0015] The multi-machine collaborative planning module utilizes advanced distributed consensus algorithms and multi-agent reinforcement learning techniques to achieve efficient task allocation and path coordination in a multi-machine environment. This module designs a distributed decision-making framework that allows each machine to independently make local optimization decisions while maintaining communication, while ensuring that the overall system goal is achieved. Specifically, the module uses consensus algorithms, such as Raft or Paxos, to ensure information synchronization and decision consistency across machines, enabling stable operation even in the face of network delays or communication interruptions. Furthermore, using multi-agent reinforcement learning, each machine learns the optimal strategy based on its own observations and those of its peers to maximize overall performance and reduce task execution time. For example, the collaborative learning rate is set to 0.01, the discount factor is set to 0.9, and the exploration rate is gradually reduced from 1.0 to 0.1 to achieve a balance between exploration and exploitation. By applying these techniques, the multi-machine collaborative planning module not only improves the efficiency of task allocation and the collaborative nature of path planning, but also enhances the adaptability of the machine group to dynamic environmental changes.

[0016] Preferably, the user interaction and remote control module enables the operator to easily monitor the real-time status of the machine, including position, path and environmental perception results, by providing an intuitive graphical user interface (GUI). The module supports manual intervention, allowing the operator to adjust the machine's motion path or set target points through a keyboard, mouse or dedicated controller in an emergency or specific task requirement. At the same time, it also supports remote command input, and the operator can send instructions such as movement, stop, obstacle avoidance and task switching from the control center through a network connection. The system also has real-time feedback and logging functions to record all manual interventions of the operator and the operating status of the machine, which is convenient for subsequent analysis and debugging. In terms of security, the module integrates multi-level security authentication, such as passwords, biometrics or digital certificates, to ensure that only authorized users can control it.

[0017] Preferably, the system self-assessment and learning module is equipped with a data acquisition unit that can collect the machine's operating data at a frequency of once per second, including key parameters such as speed, acceleration, steering angle and path deviation. Using this data, the machine learning algorithm in the module, such as the gradient boosting decision tree (GBDT) or support vector machine (SVM), continuously learns and identifies patterns of efficient path planning, and automatically adjusts the parameters of the path optimization algorithm, such as adjusting the weight of the heuristic function or redefining the obstacle avoidance strategy. In addition, the module also integrates a feedback loop mechanism to evaluate the effectiveness of the planning strategy based on the actual operation results, and optimize the algorithm accordingly, such as fine-tuning the search depth of the A* algorithm or the update frequency of the D*Lite algorithm by analyzing the time and success rate of path execution. In this way, the system self-assessment and learning module not only improves the autonomous learning ability of the machine motion planning, but also significantly improves the adaptability and efficiency of the planning strategy.

[0018] The technical solution of this system forms a complete intelligent machine motion planning solution through the organic combination of the above modules, which can adapt to various dynamic environments and improve the autonomy and intelligence level of the machine.

[0019] By adopting the above technical solution, the present invention can bring the following three key technical effects:

[0020] 1. Highly Accurate Real-Time Environmental Perception: By integrating multiple sensor sources and advanced data preprocessing techniques, the system achieves precise perception of the machine's surroundings. Specifically, lidar and stereo cameras provide precise distance and shape information, ultrasonic and infrared sensors detect obstacles at close range, and force sensors sense direct contact between the machine and the environment. After synchronization and noise reduction, this data is then processed by the Environmental Perception and Modeling module using SLAM technology to generate a highly accurate 3D environmental model, providing a solid foundation for the machine's decision-making.

[0021] 2. Dynamic Path Planning and Obstacle Avoidance: The deep learning-based predictive analysis module predicts the movement trajectory of obstacles, while the real-time path optimization module rapidly calculates the optimal path based on this prediction information and the machine's kinematic and dynamic constraints. Specifically, the system uses a modified A* or D*Lite algorithm, combined with the output of the machine learning model, to adjust the path in real time, ensuring the machine's safe and efficient navigation in complex and dynamic environments. The emergency obstacle avoidance and safety assurance module further enhances system safety, rapidly generating obstacle avoidance strategies when potential collision risks are detected to prevent accidents.

[0022] 3. Intelligent Self-Learning and Collaborative Operation Capabilities: The system's self-assessment and learning module collects machine motion data and environmental feedback, utilizing machine learning algorithms to continuously optimize planning strategies and enhance the system's autonomous learning capabilities. In a multi-machine environment, the multi-machine collaborative planning module utilizes a distributed consensus algorithm and multi-agent reinforcement learning to achieve effective task allocation and path coordination between machines, improving overall operational efficiency. Specifically, the system dynamically adjusts task allocation based on each machine's task completion status and environmental changes, optimizing resource utilization, reducing redundant work, and enabling efficient collaborative operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is the system block diagram of the present invention

[0024] Figure 2 This is the flow chart of the dynamic obstacle detection and classification module of the present invention

[0025] Figure 3 This is the flow chart of the real-time path optimization module of the present invention DETAILED DESCRIPTION

[0026] The following is a combination of the embodiments of the present invention Figures 1 to 3 The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] This invention provides a real-time motion planning system based on dynamic environment perception. This system adopts a series of innovative technical solutions to achieve efficient and safe motion planning in response to the complexity of dynamic environments. The following are the specific technical solutions of this system:

[0028] LiDAR: Rotating or solid-state; Measurement range: 0.1 to 100 meters; Resolution: 360° horizontally, 5 to 30° vertically; Data update rate: 10 to 20 Hz. LiDAR is a workhorse sensor for environmental mapping and obstacle detection. It emits and receives laser pulses to generate high-resolution point cloud data, providing machines with detailed spatial maps.

[0029] Stereo camera: Resolution: at least 1920x1080 pixels; Field of view (FOV): at least 55° horizontally and at least 45° vertically; Depth measurement accuracy: Centimeter-level stereo cameras provide depth perception capabilities, measure the three-dimensional shape and distance of objects by simulating human binocular vision, identify objects in the environment and assess their size.

[0030] Ultrasonic sensor: Measuring range: 0.02 meters to 3 meters; Angular resolution: 30° to 60°; Response time: less than 10 milliseconds; Ultrasonic sensors are responsible for close-range obstacle detection. They detect and measure the distance between the machine and obstacles by emitting ultrasonic waves and measuring the time of the echo.

[0031] Infrared sensor: Detection wavelength range: 700nm to 1000nm; temperature measurement range: -20℃ to 120℃; accuracy: ±0.5℃ to ±1℃; infrared sensors can detect temperature changes and heat sources in the environment, assisting the machine in night or indoor navigation under different lighting conditions.

[0032] Force sensor: Measuring range: 0 to 100 Newtons; Resolution: 0.1 Newtons; Response time: less than 5 milliseconds; The force sensor is installed on the contact surface of the machine and can measure and feedback the force when the machine contacts the environment, providing tactile perception for fine operations.

[0033] Noise Reduction: A Kalman filter, an optimized recursive filter, is used with a noise covariance parameter set to 0.01 and a state covariance parameter set to 1. This setting enables the filter to effectively remove high-frequency noise from the sensor data while preserving important signal features at an update frequency of 0.1 seconds.

[0034] Normalization: To eliminate the differences in the dimensions and scales of sensor data, the preprocessing module normalizes the data. For example, if the original data ranges from 0 to 100, the normalization process converts it to a range of 0 to 1.

[0035] Data synchronization: Ensuring data temporal consistency is a prerequisite for accurate multi-sensor perception. The preprocessing module utilizes a global timestamp system to achieve microsecond-level data synchronization accuracy. This means that even in high-speed or dynamically changing environments, data from different sensors remains synchronized, ensuring the machine's accurate understanding of the environment.

[0036] Data fusion: Data fusion combines information from different sensors to form a more comprehensive and consistent environmental perception model. The preprocessing module uses a weighted average method for data fusion. LiDAR may be assigned a weight of 0.6 due to its high precision, while stereo cameras may be assigned a weight of 0.3 due to their sensitivity to lighting conditions.

[0037] Computer Vision Application: Using computer vision technology, the convolutional neural network (CNN) in the module processes images captured by the stereo camera and identifies objects in the environment. The CNN input image resolution is 1920x1080 pixels and the frame rate is 30 frames per second, ensuring high definition and real-time performance.

[0038] SLAM technology implementation: The module uses SLAM technology for real-time positioning and map construction. Taking visual SLAM as an example, the feature extraction algorithm used can detect at least 100 feature points in each frame. The descriptors of these feature points are 256 bytes long and are used for tracking and matching.

[0039] High-precision 3D model generation: Combining data fusion and SLAM technology, the module generates 3D models with centimeter-level accuracy. The model is updated 10 times per second, ensuring real-time capture of environmental changes. Furthermore, the model's vertical accuracy error is controlled within 0.01 meters and its horizontal accuracy error is controlled within 0.002 meters.

[0040] Module parameter settings: In the SLAM algorithm, keyframe selection is based on the significance of environmental changes, with a change threshold of 5%. The map resolution is set to 1 meter per pixel to balance computational load and map accuracy. Feature extraction thresholds, such as the corner response threshold in the FAST algorithm, are set to 20 to optimize feature point detection speed and quality.

[0041] Using convolutional neural networks (CNN) for efficient and accurate real-time processing, this module uses a lightweight CNN model to ensure synchronization with the machine's motion at a processing speed of at least 30 frames per second. The model training uses the cross entropy loss function. To optimize category distinction, the formula is expressed as: Among them, C represents the total number of categories, y o,c is the one-hot encoding of the true label, y p,cis the probability predicted by the model. The model uses the Adam optimizer with a learning rate set to 0.001 and a batch size of 32 to achieve fast convergence and high generalization. After training, the model achieved an accuracy of over 95% on the test set and completed the inference process for a single frame image within 200 seconds. In addition, for tracking dynamic obstacles, the module uses deep learning-based tracking algorithms such as YOLO or SSD, which are capable of handling fast-moving obstacles and providing their motion trajectory and speed information. Through this technical solution that combines deep learning, real-time processing, and adaptive adjustment, the dynamic obstacle detection and classification module provides intelligent machines with a powerful environmental perception capability, ensuring efficient motion planning and safe navigation in complex dynamic environments.

[0042] The deep learning-based predictive analysis module uses recurrent neural networks (RNNs) and reinforcement learning techniques to predict and analyze obstacle trajectories and environmental trends. This module utilizes a long short-term memory (LSTM) network, a special type of RNN, with the following parameters: Number of hidden layer neurons: 256, providing sufficient capacity for the network to capture complex patterns in time series data. Learning rate: 0.001, adjusted using the Adam optimizer, to ensure rapid convergence during training. Time step: Each prediction cycle considers movement within the next 5 seconds, with a temporal resolution of 0.1 seconds, providing sufficiently detailed predictions. Discount factor: 0.95, balancing immediate and future rewards, allowing predictions to prioritize the impact of long-term behavior. The parameters of the Deep Q-Network (DQN) integrated in this module are as follows: Network architecture: Constructed from two convolutional layers for spatial feature extraction and two fully connected layers for decision-making. Batch size: 32, ensuring efficiency and stability of each training update. Exploration rate: Initially set to 1.0 and gradually reduced to 0.1, it controls the balance between the machine's exploration of the environment and its use of known strategies. The predictive analysis module accurately predicts the future movement trajectories of obstacles, providing critical forward-looking information for the machine's real-time motion planning, significantly improving the machine's autonomy and safety in dynamic environments.

[0043] The real-time path optimization module uses a modified A* algorithm or D*Lite algorithm through carefully set parameters to achieve fast and efficient motion path search and optimization. Specifically, the module uses Manhattan distance as a heuristic function and sets the search depth to 20 steps to ensure that possible paths can be fully explored in complex environments. For the D*Lite algorithm, the path update frequency is set to 10 times per second to adapt to rapid changes in the environment. In the A* algorithm, the node expansion rate is controlled at 80% to balance the breadth and depth of the search and prevent the search process from falling into a local optimal solution. In addition, the module utilizes the parallel processing capabilities of multi-core processors to distribute path search tasks to 4 cores for execution, significantly improving computational efficiency. The optimized configuration of these parameters enables the real-time path optimization module to meet the real-time requirements of machine motion planning while ensuring the quality of path optimization.

[0044] Ensure that path planning not only considers environmental factors, but also the physical characteristics of the machine. The module carefully analyzes parameters such as the machine's joint angle limits, maximum speed, and acceleration. For example, it sets the rotation range of each joint of the machine to between -180° and 180°, the maximum linear speed to 1.5 m / s, and the maximum acceleration to 0.5 m / s2. Through strict verification of these kinematic parameters, the module automatically adjusts the path points to avoid exceeding the machine's movement capabilities. At the same time, it uses a dynamic model to evaluate the energy consumption and stability of the path to ensure the safety and efficiency of the machine when executing the path. In addition, the module also integrates a collision detection algorithm to monitor potential collision risks on the path in real time and make necessary adjustments to the obstacle avoidance strategy to achieve smooth, coherent and physically compliant motion planning.

[0045] The emergency obstacle avoidance and safety assurance module integrates advanced collision detection algorithms, such as geometric model-based collision detection, which can monitor the distance between the machine and obstacles in real time. The safety distance threshold is set at 0.5 meters. Once the distance between the machine and the obstacle is detected to be less than this threshold, the system triggers an alarm. At the same time, the module uses a safety distance assessment mechanism to evaluate the current speed of the machine and the movement state of the obstacle to calculate the minimum safe distance. When a potential collision risk is detected, the module can quickly generate obstacle avoidance strategies, such as slowing down, stopping, or changing the direction of movement, and by executing these strategies, ensure the safety of the machine and the surrounding environment. In addition, the module also has self-learning capabilities, which can continuously optimize the obstacle avoidance strategy based on historical obstacle avoidance data, thereby improving the safety and reliability of the machine in complex environments.

[0046] The multi-machine collaborative planning module designs a distributed decision-making framework that allows each machine to independently make local optimization decisions while maintaining communication, while ensuring that the global goals of the entire system are achieved. Specifically, the module uses consensus algorithms, such as the Raft or Paxos protocols, to ensure information synchronization and decision consistency among machines, maintaining stable operation even in the face of network delays or communication interruptions. At the same time, using multi-agent reinforcement learning, each machine can learn the optimal strategy based on its own observations and those of its peers to maximize overall efficiency and reduce task execution time. For example, the collaborative learning rate is set to 0.01, the discount factor is set to 0.9, and the exploration rate is gradually reduced from 1.0 to 0.1 to achieve a balance between exploration and exploitation. Through the application of these technologies, the multi-machine collaborative planning module not only improves the efficiency of task allocation and the coordination of path planning, but also enhances the adaptability of the machine group to dynamic environmental changes.

[0047] The user interaction and remote control module provides a powerful operator interaction experience through an intuitive graphical user interface (GUI). This interface allows operators to input commands through a high-precision touchscreen or dedicated console, supporting a display resolution of at least 1024×768, ensuring clear visualization of maps and machine status. In manual control mode, operators can adjust the machine's motion path or directly control machine movement using mouse or keyboard input at an update rate of at least 10 times per second. Remote command input uses secure encryption protocols such as TLS 1.2 to ensure secure command transmission and supports a command transmission rate of at least 50 times per second, ensuring real-time performance. The system feedback mechanism responds to all operator inputs within 200 milliseconds and displays real-time feedback on the interface. A logging function records all operator interventions and machine status changes at least once per second, facilitating subsequent analysis and system optimization. Furthermore, security and permission management features utilize a multi-level authentication mechanism, including passwords, biometrics (such as fingerprint or facial recognition), and digital certificates, ensuring only authorized users have control, thereby enhancing system safety and reliability. The careful setting of these parameters enables the user interaction and remote control module to not only provide efficient operator interaction capabilities but also ensure the flexibility and controllability of the system.

[0048] The system's self-assessment and learning module continuously optimizes its planning strategy through precise data collection and advanced machine learning algorithms. The module's data acquisition unit records the machine's motion status in real time at a rate of 50 times per second, including speed (0.1 m / s2 to 2 m / s2), acceleration (0.01 m / s2 to 1 m / s2), steering angle (-180° to 180°), and path deviation (with an accuracy of 1 cm). Leveraging this data, the module's machine learning algorithms, such as the Gradient Boosted Decision Tree (GBDT), employ a learning rate of 0.01 and a tree depth of 0.5 to continuously learn and identify efficient path planning patterns. Simultaneously, a Support Vector Machine (SVM) algorithm, with a regularization parameter of 0.001 and a Radial Basis Function (RBF) kernel, performs classification and regression analysis on environmental feedback. The module also integrates a feedback loop mechanism that dynamically adjusts the A* algorithm's search depth (ranging from 5 to 20 steps) and the D*Lite algorithm's update frequency (every 0.1 to 1 second) based on metrics such as average path execution time and success rate. Through the precise setting of these parameters and continuous learning of the algorithm, the system self-assessment and learning module significantly improves the machine's autonomous learning ability and the efficiency and adaptability of motion planning.

[0049] Through these steps, the real-time motion planning system based on dynamic environment perception of the present invention realizes the entire process from environmental perception to path planning, obstacle avoidance, multi-machine collaborative operation, and then to user interaction and system self-optimization, ensuring the efficient and safe operation of the machine in a complex dynamic environment.

[0050] 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 present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time motion planning system based on dynamic environment perception, characterized in that: It includes the following modules: multi-source sensor integration module, advanced data preprocessing module, environmental perception and modeling module, dynamic obstacle detection and classification module, deep learning prediction and analysis module, real-time path optimization module, kinematic and dynamic constraint processing module, emergency obstacle avoidance and safety assurance module, multi-machine collaborative planning module, user interaction and remote control module; the multi-source sensor integration module collects the visual, distance, sound and tactile data of the machine's surrounding environment in real time to provide the system with comprehensive perception information; the advanced data preprocessing module reduces noise, standardizes and synchronizes the received raw data to ensure the accuracy and consistency of the data; the environmental perception and modeling module uses the fused data to generate a high-precision three-dimensional model of the machine's operating environment to provide detailed spatial information for subsequent planning; on this basis, the dynamic obstacle detection and classification module identifies and classifies static and dynamic obstacles in the environment in real time to provide key information for obstacle avoidance and path planning; the prediction and analysis module based on deep learning The movement trajectory of obstacles and environmental change trends are further predicted to provide forward-looking information for the system. The real-time path optimization module dynamically adjusts the optimal movement path of the machine according to the environmental model and prediction results to ensure the real-time and adaptability of path planning. The kinematic and dynamic constraint processing module ensures that the planned path meets the physical and operational limitations of the machine and ensures the feasibility and safety of machine movement. The emergency obstacle avoidance and safety assurance module quickly generates obstacle avoidance strategies when potential collisions are detected to further improve the safety of the system. The multi-machine collaborative planning module uses a distributed algorithm to achieve task allocation and path coordination among multiple machines, optimize resource utilization and improve work efficiency. The user interaction and remote control module provides the operator with the ability of manual intervention and remote command input to enhance the flexibility and controllability of the system. The system self-assessment and learning module continuously optimizes planning strategies and improves planning efficiency through machine learning algorithms, so that the system can automatically improve based on experience to achieve long-term self-optimization and performance improvement.

2. The system according to claim 1, characterized in that The multi-source sensor integration module includes but is not limited to a laser radar (LIDAR), a stereo camera, an ultrasonic sensor, an infrared sensor and a force sensor.

3. The system according to claim 1, characterized in that The advanced data preprocessing module uses filtering algorithms and data calibration techniques to improve data quality.

4. The system according to claim 1, characterized in that The environment perception and modeling module uses computer vision and SLAM technology to perform real-time three-dimensional reconstruction of the environment.

5. The system according to claim 1, characterized in that The dynamic obstacle detection and classification module adopts image recognition and object tracking algorithms based on deep learning.

6. The system according to claim 1, characterized in that The deep learning-based forecasting analysis module combines time series analysis and reinforcement learning for forecasting.

7. The system according to claim 1, characterized in that The real-time path optimization module uses an improved A* or D*Lite algorithm to perform fast path search.

8. The system according to claim 1, characterized in that The kinematic and dynamic constraint processing module takes into account the joint limits, velocities and accelerations of the machine.

9. The system according to claim 1, characterized in that The emergency obstacle avoidance and safety assurance module integrates collision detection algorithm and safety distance assessment.

10. The system according to claim 1, characterized in that The multi-machine collaborative planning module uses an optimization algorithm based on graph theory to perform task allocation and path coordination.

Images