Device with obstacle avoidance decision function

Abstract

The application discloses a device with obstacle avoidance decision function, and belongs to the field of non-electric variable control or regulation system. The application is used to solve the technical problems of high power consumption, large delay and being easily disturbed by self motion and environmental light of the existing obstacle avoidance scheme. The application comprises a modulated light source, an event-based sensor and a processing unit. The modulated light source generates light at a specific frequency and forms a three-dimensional space. The event-based sensor senses the change of light intensity in the three-dimensional space and outputs an event stream. The processing unit is configured to trigger an obstacle avoidance decision process if an object contour is detected according to the event stream. The application can be used for real-time obstacle avoidance of mobile robots, unmanned aerial vehicles or intelligent vehicles, and has the characteristics of low power consumption, small delay, high robustness and low cost.

Description

Technical Field

[0001] This invention relates to the field of control or regulation systems for non-electrical variables, and more specifically to a device with obstacle avoidance decision-making function. Background Technology

[0002] With the rapid development of autonomous mobile robots, drones, and intelligent vehicles, obstacle avoidance systems, as a core module ensuring their safe and autonomous operation, have become a research hotspot in both academia and industry. The key to obstacle avoidance systems lies in quickly and accurately perceiving obstacle information in the path, providing reliable input for path planning and decision-making control. Currently, mainstream obstacle avoidance detection schemes are mainly based on the following types of sensor technologies:

[0003] The first type is ultrasonic sensors. These sensors emit ultrasonic waves and detect the reflected waves, using the speed of sound to calculate the distance to obstacles. Ultrasonic technology is mature and the hardware cost is low, but its detection accuracy is poor, it is easily affected by sound-absorbing materials or environmental noise, and its detection frequency is low, making it difficult to meet the real-time obstacle avoidance requirements in high-speed motion scenarios.

[0004] The second category is lidar and structured light cameras. This type of solution obtains high-precision depth information by emitting lasers or coded light and utilizing time-of-flight or triangulation principles. Although the detection accuracy is high, lidar is large and expensive, and depth cameras have extremely high requirements for processor performance, resulting in high system hardware costs and high power consumption, which is not conducive to the miniaturization of equipment and cost control.

[0005] The third category is traditional infrared / RGB cameras. This type of solution continuously acquires environmental image frames and combines them with deep learning algorithms (such as convolutional neural networks) for obstacle detection and semantic segmentation. However, this method requires processing massive amounts of pixel data, resulting in high overall system power consumption (typically in the watt range) and significant processing latency, making it difficult to deploy long-term in micro-drones or low-power wearable devices. Furthermore, frame-based algorithms are sensitive to lighting conditions; in strong light overexposure, low light noise, or complex dynamic backgrounds, the recognition accuracy is prone to fluctuations.

[0006] The first and second types of solutions mentioned above can be further classified as traditional obstacle avoidance methods, while the third type of solution can also be regarded as an obstacle avoidance method based on deep learning.

[0007] In recent years, visual perception methods based on event cameras have gradually attracted attention. Event cameras (also known as dynamic vision sensors, DVS) have advantages such as microsecond-level response time, high dynamic range (up to 120dB or more), and low power consumption, enabling them to asynchronously sense changes in light intensity in a scene and output sparse event streams. Based on the unique operating characteristics of event cameras, various visual information systems have been proposed in the industry.

[0008] In light of this, obstacle avoidance methods based on processing event streams have been developed in this field, as well as obstacle avoidance methods based on sensor fusion. These different types of obstacle avoidance schemes can be referenced in prior art 1 and prior art 2 (obstacle avoidance schemes that fuse RGB and event streams).

[0009] Existing technology 1: Wang Jialiang, Dong Kai, Gu Zhaojun, Chen Hui, Han Qiang. A review of obstacle avoidance methods using visual sensors for small unmanned aerial vehicles [J]. Journal of Xidian University, 2025, 52(1): 60-79.

[0010] Existing technology 2: Cai Zhihao, Chen Wenjun, Zhao Jiang, et al. Target detection and obstacle avoidance of UAV based on dynamic vision sensor [J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(1): 144-153.

[0011] Prior art 3: CN121349303A, a spatial visual information system and perception method.

[0012] Existing technology 3 demonstrates significant advantages in low power consumption, low latency, and high-precision target mask generation, and is mainly applicable to human-computer interaction scenarios such as augmented reality (AR) and virtual reality (VR) for obtaining accurate contours of interactive targets such as hands. Although it utilizes an event camera, existing technology 3 does not make any indication of obstacle avoidance for autonomous devices.

[0013] In the field of obstacle avoidance technology, event cameras are commonly used to directly observe surrounding objects (potential obstacles), just like RGB cameras, which is a common practice in this field. However, whether using event cameras alone or sensor fusion obstacle avoidance solutions, complex algorithms and powerful computing capabilities are usually required, which poses a significant challenge to the battery life of autonomous devices.

[0014] Therefore, how to achieve obstacle avoidance functionality for autonomous devices at the cost of extremely low energy consumption remains a major challenge in this field. Summary of the Invention

[0015] To alleviate or partially alleviate the above-mentioned technical problems, the solution of the present invention is as follows:

[0016] This invention discloses a device with obstacle avoidance decision-making function, comprising:

[0017] A light source that generates light at a modulated frequency and forms a three-dimensional space;

[0018] An event-based sensor is used to sense changes in light intensity in the three-dimensional space; and,

[0019] The processing unit is configured to trigger an obstacle avoidance decision process if an object outline is detected based on the event stream output by the event-based sensor.

[0020] In one embodiment, the three-dimensional space is a space whose cross-sectional area gradually increases along the direction of light propagation; the light source is a laser light source or an LED light source; and the device is configured to allow adjustment of the effective detection area of ​​the device by adjusting the emission power of the light source.

[0021] In one embodiment, the device with obstacle avoidance decision-making function further includes:

[0022] A chip that runs a spiking neural network;

[0023] The spiking neural network is configured to receive the event stream output by the event-based sensor and perform frequency-selective filtering to extract valid events that match the modulation frequency.

[0024] The spiking neural network includes a LIF neuron layer, which is used to filter out events other than valid events that match the modulation frequency.

[0025] In one type of embodiment, the device is configured to perform:

[0026] Accumulate and count valid events with the same coordinates over time, and use multiple accumulated counts to generate the object outline;

[0027] The accumulated count decays over time or is cleared directly at regular intervals.

[0028] In one embodiment, the obstacle avoidance decision process adopts an event-driven approach, triggering the obstacle avoidance decision only when an object outline is detected.

[0029] In one type of embodiment, it further includes:

[0030] The coordinate mapping module is used to map the detected object contour from the sensor coordinate system to the motion control coordinate system of the device.

[0031] In one embodiment, the obstacle avoidance decision-making process includes a hierarchical decision architecture, which includes a behavior decision layer and a motion control layer. The behavior decision layer determines a behavior pattern based on object contour information, and the motion control layer generates control commands based on the behavior pattern.

[0032] In one embodiment, the behavior mode includes one or more of the following modes: normal driving mode, deceleration following mode, emergency obstacle avoidance mode, or stationary standby mode.

[0033] In one embodiment, the device is further configured to adjust the emission power or modulation frequency of the light source according to the state of the obstacle avoidance decision process.

[0034] In one embodiment, the device is a mobile robot, a drone, or a smart vehicle.

[0035] The technical solution of this invention has one or more of the following beneficial technical effects:

[0036] (1) It has high robustness. The present invention forms a three-dimensional light field by modulating the frequency and combines it with a frequency selective filtering mechanism to effectively suppress background flow field noise and ambient light flicker interference generated by the sensor's own motion. Even under conditions of high-speed motion of autonomous equipment such as vehicles and complex lighting, it can still stably extract obstacle reflection signals, significantly improving dynamic environment adaptability.

[0037] (2) Achieving low power consumption and low latency: Compared to traditional vision solutions that require a large amount of computing power for direct observation, this invention uses a method of actively constructing a three-dimensional light field to observe surrounding objects, greatly simplifying computing power consumption. Adopting an event-driven architecture, the system is activated only when an obstacle enters the light field (in a specific embodiment, the overall power consumption can be controlled below 5mW). The microsecond-level response and asynchronous processing mechanism of the event camera enable obstacle perception and mask generation rates of hundreds of hertz, significantly reducing obstacle avoidance decision latency.

[0038] (3) Achieving low cost and high precision, this invention directly generates obstacle masks through spatiotemporal superposition within the event domain, achieving pixel-level contour extraction accuracy. It does not require high-performance processors and complex optical devices, reducing hardware costs compared to traditional solutions (a specific example shows a reduction of approximately 70%), and is compact in size, making it easy to integrate into various mobile platforms.

[0039] Furthermore, other beneficial effects of the present invention will be mentioned in the specific embodiments. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating the obstacle avoidance principle of one embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the application scenario of the present invention in intelligent robots. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0043] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order.

[0044] This invention provides a device with obstacle avoidance decision-making capabilities, suitable for scenarios such as mobile robots, drones, and intelligent vehicles that require rapid, low-power obstacle perception and spatial boundary detection during movement. The technical solution of this invention will be described in detail below with reference to the accompanying drawings.

[0045] Furthermore, the technical means in prior art 3 are incorporated herein by reference as a supplementary description of other unresolved details.

[0046] Figure 1 This is a schematic diagram illustrating the obstacle avoidance principle of one embodiment of the present invention. As shown, the obstacle avoidance system of the device of the present invention includes a modulated light source emitter and an event-based sensor. The modulated light source emitter generates light at a specific modulation frequency, forming a three-dimensional light field region in the forward direction of the robot or vehicle. The three-dimensional light field region can be a cone-shaped space with a gradually increasing cross-sectional area along the direction of light propagation, or other three-dimensional spaces, which can be adjusted according to the distance range and field of view requirements for obstacle avoidance detection. Preferably, the modulated light source emitter uses an infrared band (e.g., 850nm) laser light source to reduce the interference of ambient visible light on the operation of the obstacle avoidance system, while avoiding the impact of visible light flicker on the user experience. The modulated light source emitter is driven by a pulse signal, and the frequency and duty cycle of the driving signal can be set according to the actual application scenario, for example, set to 300Hz and 50% duty cycle, to generate a stable modulated light field.

[0047] Event-based sensors are used to sense changes in light intensity within the three-dimensional light field region. In one embodiment, the event-based sensor is a dynamic vision sensor (DVS), which asynchronously detects the logarithmic change in light intensity at each pixel location and immediately outputs an event when the change exceeds a preset threshold. Each event data includes the pixel coordinates (x, y) where the event occurred, a microsecond-level timestamp t, and the polarity p∈{+1,-1} representing the direction of the light intensity change. The asynchronous event stream composed of these events can be represented as E={e_i=(x i ,y i ,t i ,p iLet the region be a set of pixels, i = 1, ..., N, where N is a positive integer. When no obstacle enters the region, the pixels of the dynamic vision sensor sense a stable changing background, which may not generate any events or only a small number of random, sporadic events caused by ambient light noise. In another embodiment, the event-based sensor may also be a hybrid vision sensor (such as DAVIS) that simultaneously includes DVS pixels and active pixel sensor (APS) pixels. To ensure that the sensor can effectively sense changes in light intensity within the three-dimensional light field region, the field of view of the event-based sensor should completely cover or at least partially cover the three-dimensional light field region formed by the modulated light source emitter.

[0048] Figure 2 This is a schematic diagram illustrating the application scenario of this invention in an intelligent robot. When the robot moves along its original trajectory, if an obstacle is present, the modulated light emitted by the modulated light source emitter is reflected at the obstacle. As shown in the figure, the reflected modulated light enters the field of view of the event-based sensor, is sensed by the sensor, and generates a corresponding event. Specifically, when an obstacle appears in the three-dimensional light field region, it obstructs or reflects the modulated light, causing rapid changes in local light intensity. These changes are captured by the event-based sensor and output as a sparse event stream. It should be noted that the modulated light not obstructed by obstacles in the three-dimensional light field region propagates outwards. Although other distant obstacles may scatter or reflect the modulated light, the intensity of the reflected light is greatly reduced due to the distance, resulting in a weak change in light intensity that is difficult for the event-based sensor to effectively capture. This improves the robot's obstacle avoidance accuracy. The robot only triggers obstacle avoidance when the obstacle is a certain distance away, and distant obstacles do not affect the robot's obstacle decision.

[0049] The obstacle avoidance system of the device of the present invention also includes a chip running a spiking neural network (SNN). This chip can be a dedicated neuromorphic chip or an SNN software model running on a general-purpose processor. The spiking neural network is configured to receive an event stream from event-based sensor outputs and perform frequency-selective filtering to extract valid events that match the modulation frequency of the light source. In a preferred embodiment, the spiking neural network includes a layer of leak-integral-fire (LIF) neurons. The membrane potential dynamics of the LIF neurons satisfy a first-order differential equation: , where τ m Let R be the membrane time constant, R be the membrane resistance, and I(t) be the input current. In frequency domain analysis, the LIF neuron is equivalent to a first-order low-pass filter, and its transfer function is: By carefully setting τ mBy setting the membrane time constant of the LIF neuron to match the modulation frequency of the light source, the LIF neuron layer can effectively suppress events that are mismatched with the modulation frequency, including interference events such as ambient light noise and background flow field noise generated by the sensor's own motion, while retaining valid events that are consistent with the modulation frequency. For example, when the light source modulation frequency is 300Hz, the membrane time constant of the LIF neuron can be set to a corresponding value, so that it has the best response to signals near 300Hz, while attenuating other frequency components.

[0050] It should be noted that using a pulse neural network for frequency-selective filtering to extract valid events matching the modulation frequency of the light source is only one preferred implementation of the present invention. Those skilled in the art will understand that any signal processing method capable of extracting valid events matching a specific modulation frequency from an event stream falls within the scope of the present invention. For example, in other embodiments, a digital bandpass filter can be used to perform frequency domain filtering on the time series of the event stream, or an analog lock-in amplifier circuit can be used to demodulate the event trigger signal, or a conventional digital signal processor (DSP) can run a frequency domain analysis algorithm to extract events with specific frequency components. Any alternative technology that can separate valid signals from noise from an event stream based on frequency characteristics, provided its core function is substantially the same as the frequency-selective filtering effect achieved by the present invention, should be considered an equivalent replacement for the present invention.

[0051] In one embodiment, a filter is placed in front of the DVS lens, which only allows light of the wavelength emitted by the modulated light source to pass through the filter. The advantage of this embodiment is that it can easily and efficiently filter out more interference from irrelevant wavelengths of light (such as those caused by moving objects in the background) on the DVS light intensity information changes using only physical means, avoiding more and more complex noise reduction schemes.

[0052] The filtered event stream has removed most of the noise interference, and obstacle imaging extraction is performed next. This invention generates a real-time mask of obstacles within the event domain through spatiotemporal overlay. Specifically, a projection buffer with the same resolution as the event-based sensor is constructed, where the value of the buffer unit corresponding to pixel coordinates (x, y) is I. proj (x, y, t). For each filtered valid event (x... i , y i , t i This process updates the value of the corresponding buffer cell in the projection buffer. This process can be formally represented as... The summation is performed on the filtered set of valid events. Over time, events continuously generated by obstacles in the 3D light field accumulate in the projection buffer, forming a clear event density distribution map that directly corresponds to the geometric contour of the obstacle in the sensor's field of view. In one implementation, the values ​​in the projection buffer can decay over time, for example, exponentially. Newly arriving events increase the value of the buffer unit at their corresponding spatial location, while the contribution of older events decays over time, allowing the mask to reflect real-time changes in the obstacle's position. In another implementation, all values ​​in the projection buffer can be periodically cleared to avoid the cumulative effect of historical events.

[0053] The generated obstacle mask is located in the sensor coordinate system. To apply it to actual obstacle avoidance decisions, it needs to be mapped to the robot's motion control coordinate system. This coordinate mapping process can be achieved using calibration and transformation methods known in the art. For example, by using pre-calibrated camera intrinsic and extrinsic parameters, the pixel coordinates can be converted to coordinates in the camera coordinate system, and then combined with the robot's pose information to transform them into the robot's motion control coordinate system. After coordinate mapping, the obstacle avoidance system obtains the position and contour information of the obstacle in physical space.

[0054] It should be noted that, in this invention, "contour" is a representation of the core visual features of the geometric shape, spatial position, and external structure of the perceived obstacle, while "mask" (e.g., an event density distribution map) is a specific implementation used to represent and extract the contour. Generating a mask through event accumulation and then identifying the object contour from the mask is the preferred method of this invention, but this invention does not exclude other technical means that can directly extract the object contour from the event stream.

[0055] Based on the above obstacle information, the device of the present invention triggers an obstacle avoidance decision-making process. For example... Figure 2 As shown, when an obstacle is detected ahead of the original driving trajectory, the obstacle avoidance decision module plans a new driving trajectory to avoid the obstacle based on the obstacle's position, size, and distance information, combined with the robot's kinematic model and control objectives. For example, upon encountering an obstacle, the robot may change direction and take another driving trajectory, or follow an S-shaped movement trajectory to avoid the obstacle. Obstacle avoidance decision-making can be implemented using various known path planning algorithms, such as the artificial potential field method, the fast random search tree (RRT) algorithm, and the dynamic window method. This invention does not limit the specific obstacle avoidance decision-making algorithm; its core lies in providing the obstacle avoidance decision-making module with low-latency, high-precision obstacle perception input through the aforementioned light field encoding and event-driven perception mechanism.

[0056] It is worth noting that the obstacle avoidance decision-making process in this invention does not necessarily mean that the device will immediately move away from the obstacle in its subsequent trajectory. This often depends on various parameters (the device's current speed, the distance between it and the obstacle, etc.). This invention does not limit the specific content or form of the obstacle avoidance decision.

[0057] Specifically, the obstacle avoidance decision-making process triggered by the detection of an object's outline based on an event flow in this invention is essentially about replanning the subsequent travel path based on information such as the obstacle's position, size, and motion state, combined with the device's current pose, speed, and kinematic constraints, to achieve safe and efficient obstacle avoidance. That is, "triggering an obstacle avoidance decision" is not limited to simply commanding the device to immediately change direction or brake suddenly; it can also initiate a path replanning process: starting from its current position and aiming to avoid the detected obstacle, the device generates a new feasible path (such as a detour trajectory or an S-shaped avoidance trajectory in local path planning) and travels along this new path.

[0058] In one specific embodiment of the present invention, the device operates in an event-driven manner. When no obstacle enters the three-dimensional light field, the light source module continuously emits modulated light, but the event-based sensor does not detect significant changes in light intensity, generating almost no events or only a small amount of noise events, and the device is in a standby state with extremely low power consumption. When an obstacle enters the three-dimensional light field, the reflected light triggers the sensor to generate an event stream, and the subsequent SNN filtering, event accumulation, and mask generation modules are activated sequentially, ultimately triggering an obstacle avoidance decision. This event-driven operating mode allows the obstacle avoidance system to consume energy only when it needs to process obstacle information, and the overall power consumption can be controlled below 5mW, significantly lower than traditional frame-based visual obstacle avoidance schemes.

[0059] In another embodiment, the decision module operates at a fixed frequency, but reads the latest obstacle mask in each decision cycle. This approach is simple to implement and suitable for scenarios where latency requirements are not extremely strict.

[0060] In one embodiment, to further improve the safety and reliability of the device, the obstacle avoidance decision module can also introduce a hierarchical decision architecture. A behavior decision layer is set at the top level, responsible for determining the current behavior mode based on obstacle information and task objectives, including normal driving, deceleration following, emergency obstacle avoidance, and stationary standby. A motion control layer is set at the bottom level, responsible for translating the behavior decisions into specific motor, servo, or propeller control commands. For example, when an obstacle ahead is detected to be far away and relatively slow, the behavior decision layer selects the deceleration following mode, and the motion control layer correspondingly reduces the target speed; when the obstacle distance is detected to suddenly shorten to below the emergency threshold, the behavior decision layer immediately switches to the emergency obstacle avoidance mode, and the motion control layer executes maximum braking or rapid steering commands. This hierarchical architecture enables the device to respond in stages according to the degree of danger, ensuring safety while avoiding overreaction to minor disturbances.

[0061] In one embodiment, the obstacle avoidance decision module and the light source modulation frequency are designed in tandem. For example, when the decision module determines that an emergency obstacle avoidance state is in effect, it can temporarily increase the light source's emission power or adjust the modulation frequency to enhance the strength of the obstacle reflection signal or improve the temporal resolution of the perception, thereby providing more accurate obstacle information for emergency obstacle avoidance. Once the danger has passed, the light source returns to normal operating mode to save power. This collaborative control mechanism between the light source and the decision-making process further enhances the device's adaptability.

[0062] In one embodiment, the device of the present invention is a robotic vacuum cleaner, and the obstacle avoidance system of the device is integrated in front of the robot for real-time detection of obstacles in the travel path. In another embodiment, the device is a drone, and the obstacle avoidance system is integrated below or in front of the drone for obstacle avoidance and terrain following. In yet another embodiment, the device is an intelligent vehicle for obstacle warning and avoidance in assisted driving or autonomous driving scenarios.

[0063] In other embodiments of the present invention, various modifications and substitutions can be made to the above technical solutions. For example, in addition to infrared lasers, the light source module can also use LED light sources or other modulated light sources, as long as it can generate modulated light at a specific frequency. Besides DVS, other types of event-based visual sensors can also be used for event-based sensors. In the event filtering stage, in addition to LIF neurons in SNNs, digital bandpass filters can also be used to filter the event stream by frequency, although the latter's power consumption and latency characteristics may not be as good as the former. In the mask generation stage, in addition to direct event overlay methods, the event stream can be accumulated to form grayscale frames or event density maps, and then traditional image processing algorithms can be used to extract contours. However, this method introduces additional latency and increases computational resource consumption, making it a suboptimal choice for the present invention. In the coordinate mapping stage, in addition to the aforementioned methods, for non-flat terrain or scenes requiring 3D obstacle avoidance, more complex perspective transformations or 3D reconstruction methods can also be used.

[0064] In summary, this invention proposes a novel obstacle avoidance scheme by combining event-based sensors, modulated light sources, and spiking neural networks. This scheme actively constructs a perceptible 3D space using light field frequency encoding, achieves low-power perception with microsecond-level response using the event-driven characteristics of DVS, and suppresses background flow field noise generated by sensor motion using frequency-selective filtering of SNN. Finally, it directly generates an obstacle mask within the event domain and triggers obstacle avoidance decisions. Compared with existing technologies, this invention significantly improves environmental adaptability and robustness in high-speed motion scenarios while maintaining the advantages of low power consumption and low cost, providing a novel technical path for real-time obstacle avoidance of mobile devices. This invention is particularly suitable for applications such as robotic vacuum cleaners and hotel delivery robots (two common types of mobile robots).

[0065] To better illustrate the present invention, numerous specific details have been provided in the detailed embodiments described above. Those skilled in the art should understand that the present invention can be practiced even without certain specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art have not been described in detail, in order to highlight the spirit of the present invention.

[0066] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A device with obstacle avoidance decision-making function, characterized in that, include: A light source that generates light at a modulated frequency and forms a three-dimensional space; An event-based sensor is used to sense changes in light intensity in the three-dimensional space; as well as, The processing unit is configured to trigger an obstacle avoidance decision process if an object outline is detected based on the event stream output by the event-based sensor.

2. The device with obstacle avoidance decision-making function according to claim 1, characterized in that: The three-dimensional space is a space whose cross-sectional area gradually increases along the direction of light propagation; The light source is a laser light source or an LED light source; The device is configured to allow adjustment of the effective detection area by adjusting the emission power of the light source.

3. The device with obstacle avoidance decision-making function according to claim 1 or 2, characterized in that: The event stream output from the event-based sensor is received and frequency-selectively filtered to extract valid events that match the modulation frequency. The spiking neural network includes a LIF neuron layer, which is used to filter out events other than valid events that match the modulation frequency.

4. The device with obstacle avoidance decision-making function according to claim 3, characterized in that: The device is configured to perform: Accumulate and count valid events with the same coordinates over time, and use multiple accumulated count results to generate object contour mask information; The accumulated count decays over time or is periodically cleared.

5. The device with obstacle avoidance decision-making function according to claim 1 or 3, characterized in that: The obstacle avoidance decision-making process adopts an event-driven approach, triggering the obstacle avoidance decision only when an object outline is detected.

6. The device with obstacle avoidance decision-making function according to claim 1 or 3, characterized in that: Also includes: The coordinate mapping module is used to map the detected object contour from the sensor coordinate system to the motion control coordinate system of the device.

7. The device with obstacle avoidance decision-making function according to claim 1 or 3, characterized in that: The obstacle avoidance decision-making process includes a hierarchical decision-making architecture, which includes a behavior decision-making layer and a motion control layer. The behavior decision-making layer determines the behavior pattern based on the object contour information, and the motion control layer generates control commands based on the behavior pattern.

8. The device with obstacle avoidance decision-making function according to claim 1, characterized in that: The behavior mode includes one or more of the following modes: normal driving mode, deceleration following mode, emergency obstacle avoidance mode, or stationary standby mode.

9. The device with obstacle avoidance decision-making function according to claim 1 or 3, characterized in that, The device is also configured to adjust the emission power or modulation frequency of the light source according to the state of the obstacle avoidance decision process.

10. The device with obstacle avoidance decision-making function according to claim 1 or 3, characterized in that: The device is a mobile robot, drone, or intelligent vehicle.

Images