A non-contact photoelectric detection-enabled obstacle avoidance method for robots

Abstract

This invention discloses a non-contact photoelectric detection-enabled obstacle avoidance method for robots, relating to the field of photoelectric obstacle avoidance technology. It involves using an adjustable spectral bandwidth detection light source to emit a detection beam onto the obstacle surface and receive the returned reflected signal to obtain spectral response information reflecting material characteristics. The method determines the spectral response difference and speckle noise amplitude, calculates the noise spectral ratio, and switches the spectral bandwidth selection benchmark from maximizing the spectral response difference to maximizing the material judgment confidence under the noise spectral ratio constraint. A bandwidth working range is determined, and the detection light source is adjusted and re-detected according to this range. The collision risk level is updated based on the material judgment confidence, and the robot's obstacle avoidance path decision is output. Furthermore, the discrimination threshold is adjusted in conjunction with the roughness speckle coupling strength to suppress noise amplification under narrow-spectrum detection, thereby reducing the risk of material misjudgment in scenarios with varying surface roughness and mixed materials, and maintaining the reliability of non-contact obstacle avoidance.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic obstacle avoidance technology, and in particular to a non-contact optoelectronic detection-enabled obstacle avoidance method for robots. Background Technology

[0002] When mobile robots achieve autonomous obstacle avoidance in environments with mixed materials, it is necessary to identify the surface material of obstacles to determine the collision risk level and path avoidance priority. Non-contact photoelectric detection is the main means of obtaining obstacle material information due to its fast response speed and lack of mechanical wear. Existing solutions enhance the distinguishability between reflection signals from different material surfaces by narrowing the spectral bandwidth of the detection light source. This strategy has been repeatedly verified to be effective in controlled experiments using flat standard samples and has become a design consensus in the field.

[0003] However, in real-world operating environments, obstacle surfaces undergo wear, weathering, erosion, or processing residue, resulting in varying roughness distributions, and significant gradient changes in roughness across different regions of the same object. As the spectral bandwidth narrows, the coherence of the irradiated light increases. The interference effect between the scattered wavelets of highly coherent light after scattering on a rough surface intensifies dramatically, forming a wildly fluctuating random intensity pattern. The statistical fluctuations are deeply coupled with the surface microstructure. The amplification of speckle noise from narrow-spectrum coherent light by surface micro-undulations increases nonlinearly with narrowing spectral bandwidth and increasing surface roughness gradient. Conventional calibration procedures use standard samples with a single roughness level, eliminating the roughness gradient and thus concealing this amplification effect during calibration. It only becomes apparent after actual deployment as material misclassification. When this amplification effect exceeds a critical level, the spectral difference gain from narrow-spectrum illumination is completely offset or even surpassed by the signal-to-noise ratio loss from speckle noise. The robot misclassifies high-risk materials as low-risk, rendering obstacle avoidance path prioritization ineffective.

[0004] Therefore, how to dynamically balance the gain of spectral purity on material identification and the degradation of signal quality by speckle noise under the condition that the surface roughness gradient of obstacles is unpredictable and constantly changing, so as to enable robots to maintain reliable material determination and obstacle avoidance decision-making capabilities in multi-material mixed scenarios, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This invention discloses a non-contact photoelectric detection-enabled obstacle avoidance method for robots, comprising: A detection light source with adjustable spectral bandwidth emits a detection beam toward the surface of an obstacle, receives the reflected signal returned from the surface of the obstacle, and obtains spectral response information that reflects the material characteristics of the obstacle. Based on the spectral response information, determine the degree of difference in spectral response that characterizes the distinguishability of different materials; The amplitude of speckle noise, which characterizes signal quality degradation, is determined based on the reflected signal returned from the surface of the obstacle. The ratio of the speckle noise amplitude to the spectral response difference is calculated to obtain the noise-spectral ratio. When the noise spectrum ratio exceeds the preset discrimination threshold, it is determined that the current detection has entered the speckle noise-dominated state. The selection criterion for the spectral bandwidth is switched from maximizing the spectral response difference to maximizing the material judgment confidence under the constraint of the noise spectrum ratio, and the bandwidth working range is determined. The spectral bandwidth of the detection light source is adjusted by adjusting the bandwidth working range and the detection is re-executed to obtain the material determination confidence level. The collision risk level of the obstacle is updated according to the material determination confidence level, and the robot obstacle avoidance path decision is output.

[0006] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention addresses the contradiction between the material identification gain of narrow-spectrum photoelectric detection and the amplification of speckle noise on rough surfaces in multi-material mixed environments. By simultaneously acquiring the spectral response difference, speckle noise amplitude, and noise spectrum ratio, the spectral bandwidth selection benchmark is switched in a timely manner when speckle noise dominates. The bandwidth working range is determined by maximizing the confidence of material judgment, enabling the robot to reduce the probability of material misjudgment when the surface roughness gradient changes, and improving the reliability of collision risk level updates and obstacle avoidance path decisions. Attached Figure Description

[0007] Figure 1 This is a flowchart of a non-contact photoelectric detection-enabled obstacle avoidance method for robots according to the present invention.

[0008] Figure 2 This is a flowchart of a speckle noise-dominated state recognition method according to the present invention.

[0009] Figure 3 This is a flowchart of bandwidth working range adjustment and obstacle avoidance path output according to the present invention. Detailed Implementation

[0010] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0011] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0012] This embodiment provides a non-contact photoelectric detection-enabled obstacle avoidance method for robots, aiming to solve the problems of narrow-spectrum photoelectric detection speckle noise amplification, material misjudgment, and obstacle avoidance path priority failure caused by changes in the surface roughness of obstacles in multi-material mixed environments. In robot perception systems, the spectral bandwidth of the detection light source, the spectral response difference of the obstacle surface, and the speckle noise amplitude all affect the confidence level of material determination. This embodiment acquires spectral response information, speckle noise amplitude, and noise-spectral ratio during the detection process, and dynamically determines the bandwidth operating range when speckle noise dominates, enabling the robot to maintain reliable obstacle avoidance in rough surface and multi-material mixed scenarios. Figures 1 to 3 As shown, the method specifically includes the following steps: Step S1: A detection beam is emitted to the surface of the obstacle using a detection light source with adjustable spectral bandwidth, and the reflected signal returned from the surface of the obstacle is received to obtain spectral response information reflecting the material characteristics of the obstacle.

[0013] Configure the detection light source to make its spectral bandwidth continuously variable within a preset adjustable range, and make the wavelength coverage of the detection beam correspond to each response band; control the detection beam to illuminate the surface of the obstacle at a fixed incident orientation, and receive the reflected signals returned from the surface of the obstacle within the same detection field of view; separate the received reflected signals according to wavelength channels and mark their respective wavelengths, and output the spectral response intensity spectrum of each wavelength channel according to the response intensity of the reflected signals of each channel, as spectral response information reflecting the material characteristics of the obstacle.

[0014] In terms of structural composition, the detection light source, multi-channel spectral separation component, detection array, and processing module are connected via a synchronous trigger signal. The processing module performs normalization, feature channel selection, frequency domain separation, and subsequent discrimination and decision calculations based on the reflected light intensity of each channel obtained from synchronous sampling, and outputs the decision results to the robot motion control unit. The detection light source adopts a narrowband emitting device with adjustable spectral bandwidth. The continuous adjustment of its spectral bandwidth is achieved by adjusting the injection current, adjusting the device operating temperature, or by selecting via an external tunable filter element. The receiving end adopts a multi-channel spectral separation component to send the echo to an independent detection unit according to the wavelength channel. In one implementation, the detection light source is a narrowband laser emitter. The reflected signal returned from the obstacle surface is received by a spectral detection array. Each detection unit in the spectral detection array has narrowband filters with center wavelengths arranged at equal intervals at its front end. Each detection unit synchronously collects the reflected light intensity value after transmission through the corresponding narrowband filter at the same sampling time, and arranges them in order of wavelength channel to form the spectral response information. In another implementation, the multi-channel spectral separation component consists of a dispersive element and a linear array detector, which disperses the reflected signals of different wavelengths to different pixels of the linear array detector, and each pixel outputs the response intensity of the corresponding wavelength channel.

[0015] When a mobile robot navigates in a mixed environment of multiple materials, surfaces such as the ground, glass partitions, metal supports, fabric seats, and painted pillars may appear intermittently. The robot needs to perform non-contact material identification before approaching obstacles to plan a detour path in advance. Different materials exhibit different characteristic profiles in their reflection of light across wavelengths: metals show broad-spectrum high reflectivity, painted surfaces and plastics show absorption in the mid-wavelength range, and fabrics and porous materials show significant attenuation in the near-infrared range. Reducing the spectral bandwidth can improve the distinguishability of the aforementioned wavelength-dependent reflection characteristics, but it also enhances the coherence of the incident light, thereby exacerbating speckle on rough surfaces. To eliminate the influence of light source power fluctuations on the identification results, the reflected light power received by each channel is normalized to the incident light power of that channel as the spectral response of that channel. The responses of each channel are arranged in wavelength order to form a spectral response intensity spectrum.

[0016] The typical spectral bandwidth is 0.1 nm, which can be adjusted within the range of 0.05 to 5 nm. The value is determined based on the adjustable linewidth range given in the manufacturer's specifications for the detection light source device. The typical number of wavelength channels is 8, which can be adjusted between 4 and 16. The value is determined based on the coverage requirements of the characteristic reflection bands of common materials and the channel number specifications of the receiver. The typical wavelength coverage range of the detection beam is 400 to 1000 nm, which can be adjusted based on the characteristic reflection bands of the material group to be identified. The value is determined based on the ability of the visible to near-infrared band to distinguish common materials. The typical incident azimuth angle is 30°, which can be adjusted within the range of 15° to 45°. The value is determined based on the calibration experimental results that avoid the specular reflection main lobe and ensure the echo energy.

[0017] Step S2: Determine the degree of difference in spectral response that characterizes the distinguishability of different materials based on the spectral response information.

[0018] In each wavelength channel, the channel that is sensitive to the material's reflectivity is selected as the feature channel; the response of the surface reflection signal of different material obstacles is extracted in each feature channel to determine the direction and significance of the response deviation in each feature channel; the spectral response difference is weighted and converged based on the significance of each feature channel to obtain the spectral response difference degree that characterizes the distinguishability of the material. The greater the spectral response difference degree, the easier it is to distinguish the material under the current conditions.

[0019] The spectral response intensity spectrum primarily carries material information in the bands that exhibit clear distinctions between different materials, such as the separation between metals and paint in the mid-wavelength range, and the low reflectance characteristics of glass in specific bands. Since relying solely on a single channel is susceptible to interference from surface color and stains, multiple characteristic channels are selected for collaborative convergence. This ensures that channels with consistent response directions corroborate each other, while channels with conflicting directions suppress each other. Characteristic channels are selected based on the degree of response distinction across multiple material samples, prioritizing those with the highest value and discarding those that contribute little to material differentiation. Specifically, the selection can be based on the ratio of the dispersion of each channel's response across different material categories to its dispersion within each category; channels with higher ratios are retained as characteristic channels.

[0020] In one implementation, the processing module pre-stores reference spectra for various materials. The reference spectra are measured and normalized by the offline calibration stage for each material sample under the same incident orientation and wavelength coverage range, and then stored. The measured reflected signal intensity value of each wavelength channel is subtracted from the corresponding intensity value of the pre-stored material reference spectrum to obtain the amplitude deviation of each channel. The amplitude deviations of each channel are arranged in the channel sequence to form an amplitude deviation vector. The amplitude deviation vector is then normalized to the total energy of the detected light source by taking the Euclidean norm, and the spectral response difference is output.

[0021] The typical proportion of feature channels to the total number of channels is 50%, which can be adjusted between 30% and 80%. The value is determined based on the balance between discrimination and suppression of redundant channels in multi-material calibration samples. The weights of each feature channel are equally weighted when there is no prior knowledge, and are redistributed in the subsequent calibration according to the significance of each channel's division in the range of 0 to 1. The value is determined based on the measured contribution of each channel to material discrimination during the calibration stage.

[0022] Step S3: Determine the amplitude of speckle noise, which characterizes signal quality degradation, based on the reflected signal returned from the obstacle surface.

[0023] Frequency domain separation is performed on the reflected signals of each channel to distinguish between the low-frequency baseband component reflecting the intrinsic reflectivity of the material and the high-frequency intensity fluctuation component reflecting the scattering of surface micro-undulations. The high-frequency intensity fluctuation component is extracted, and the speckle noise amplitude corresponding to each position is arranged according to the spatial sampling position on the obstacle surface to form a spatial distribution of speckle noise amplitude. The spatial distribution is obtained by multi-point sampling within the detection field of view or continuous sampling along the surface during robot movement, with each sampling position corresponding to a set of high-frequency intensity fluctuation components. Specifically, the waveform of the reflected signal is determined, and a low-pass filter is applied to the waveform to obtain the baseline envelope signal. The waveform of the reflected signal is subtracted from the corresponding baseline envelope signal to separate the high-frequency intensity fluctuation component. The root mean square value of the high-frequency intensity fluctuation component within the sampling time window is calculated as the speckle noise amplitude of each wavelength channel. The speckle noise amplitude of each wavelength channel is weighted and averaged according to the proportion of the spectral response intensity of each channel to the total response intensity to obtain the speckle noise amplitude.

[0024] When the surface of the obstacle is rough, such as sandblasted metal, coarse ceramic, or frosted glass, the coherent light undergoes multipath interference on the microscopic undulations, and the echo exhibits granular random fluctuations in space and intensity, i.e., speckle phenomenon. The speckle does not carry intrinsic material information, but it is superimposed on the reflection profile, masking the distinguishability of the material obtained by S2; the rougher the surface and the more coherent the irradiated light, the more intense these random fluctuations are.

[0025] The amplitude of speckle noise is given by the root mean square value of the high-frequency intensity fluctuation component within the sampling time window: , In the formula, I h (n) represents the high-frequency intensity fluctuation component at the nth sampling point, in μW; Ī h σ is the mean value of the high-frequency intensity fluctuation component, in μW; N is the number of sampling points; s The value represents the speckle noise amplitude, in μW. After calculating the speckle noise amplitude for each wavelength channel, the average value is calculated by weighting the values ​​according to the proportion of each channel's spectral response intensity to the total response intensity, thus obtaining the speckle noise amplitude at the current detection position.

[0026] Furthermore, based on the spatial distribution characteristics of the speckle noise amplitude, the surface roughness gradient of the obstacle is obtained. The speckle noise amplitudes at different spatial locations on the same surface are arranged by position, and their spatial variation reflects the degree of transition in surface roughness. The ratio of the difference in speckle noise amplitude between adjacent sampling locations to the distance between adjacent sampling locations is used as the surface roughness gradient. A clear gradient transition is thus observed between welds, wear zones, and smooth painted surfaces. This surface roughness gradient is used for subsequent roughness speckle coupling strength assessment. In an optional variation, the speckle noise amplitude is read at different receiving angles to form an angle versus noise energy distribution curve. A least-squares fit is performed on the angle versus noise energy distribution curve using a mixture of Gaussian and exponential functions to obtain scattering distribution morphology parameters, including the scattering half-peak width and scattering skewness. Based on the mapping relationship between the scattering half-peak width and the root mean square height of surface roughness, and combined with the characterization of spatial non-uniformity of roughness by scattering skewness, the surface roughness gradient is calculated.

[0027] The cutoff frequency for frequency domain separation is typically 5kHz, but can be adjusted within the range of 2 to 10kHz. Its value is determined based on the calibration experiment results of stable separation of the intrinsic reflection component and the micro-fluctuation scattering component of the material in the frequency band. The number of sampling points N is typically 256, but can be adjusted between 128 and 1024. Its value is determined based on the balance between the robustness and real-time performance of the root mean square estimation.

[0028] Step S4: Calculate the ratio of the speckle noise amplitude to the spectral response difference to obtain the noise spectral ratio.

[0029] After normalizing the speckle noise amplitude to a dimensionless quantity using the same scale, it is compared with the spectral response difference to obtain the noise-spectral ratio. This ratio is used to characterize whether the degradation of signal quality by speckle noise exceeds the material identification gain brought about by reducing spectral bandwidth. The same-scale normalization of the speckle noise amplitude is achieved by mapping it to the same dimensionless scale as the spectral response difference, making the two comparable.

[0030] In mixed scenarios where smooth glass and rough concrete coexist, relying solely on material distinguishability or speckle amplitude is insufficient for decision-making. Reliability depends on the relative strength of these two factors: reducing bandwidth is beneficial when material distinguishability is dominant, while further narrowing the bandwidth will exacerbate misjudgments when speckle is predominant. The noise spectrum ratio consolidates this relative relationship into a single criterion:

[0031] In the formula, σ s * represents the speckle noise amplitude after being normalized to the same scale, dimensionless; D represents the spectral response difference, dimensionless; ρ represents the noise spectrum ratio, dimensionless; the larger ρ is, the more dominant the speckle noise and the less reliable the identification.

[0032] Reducing the spectral bandwidth enhances the coherence of the illumination light, and this enhanced coherence, along with surface roughness, amplifies the speckle pattern. The coherence length of the illumination light is calculated from the current spectral bandwidth; the narrower the bandwidth, the longer the coherence length. The roughness speckle coupling strength is evaluated based on the surface roughness gradient and the coherence length corresponding to the current spectral bandwidth. The product of the coherence length and the surface roughness gradient is calculated to obtain a dimensionless coupling factor. The square of this dimensionless coupling factor is used as the evaluation value of the roughness speckle coupling strength. A preset discrimination threshold is set as the critical value corresponding to the noise spectrum ratio when the speckle noise amplitude and the spectral response difference are equal. The stronger the coupling, the earlier conservative discrimination should be initiated. The preset discrimination threshold is adjusted based on the roughness speckle coupling strength, so that it decreases as the roughness speckle coupling strength increases. Accordingly, the adjusted preset discrimination threshold is determined by the following formula:

[0033] In the formula, T0 is the baseline discrimination threshold, which is dimensionless; κ is the roughness speckle coupling intensity, which is dimensionless; β is the down-adjustment sensitivity, which is dimensionless; and T is the adjusted preset discrimination threshold, which is dimensionless.

[0034] The benchmark discrimination threshold T0 is typically set at 0.6, and can be adjusted within the range of 0.4 to 0.8. Its value is determined based on the allowable limit of material misclassification rate under multi-material calibration samples. The sensitivity β is typically set at 0.3, and can be adjusted between 0.1 and 0.5. Its value is determined based on the balance between the magnitude of the threshold shift with the coupling strength and avoiding excessive conservatism.

[0035] Step S5: When the noise spectrum ratio exceeds the preset discrimination threshold, it is determined that the current detection has entered the speckle noise-dominated state. The selection criterion for the spectral bandwidth is switched from maximizing the spectral response difference to maximizing the material judgment confidence under the constraint of the noise spectrum ratio, and the bandwidth working range is determined.

[0036] The noise spectral ratio is compared with a preset discrimination threshold to determine whether the degradation of signal quality by speckle noise has exceeded the recognition gain brought by the spectral response difference. When the noise spectral ratio exceeds the preset discrimination threshold, the current detection is determined to have shifted from a spectral gain-dominated state to a speckle noise-dominated state. In this state, the bandwidth selection criterion aimed at maximizing the spectral response difference is removed, and the target is changed to maximizing the confidence of material determination. Under the premise that the speckle constraint is not broken, candidate values ​​of spectral line bandwidth within the adjustable range are examined, and the continuous value range that optimizes the confidence is selected to determine the bandwidth working interval. To suppress false triggering in a single frame, the percentage of frames whose noise spectral ratio exceeds the preset discrimination threshold is statistically analyzed over multiple consecutive detection cycles. When the percentage of such frames exceeds a preset proportion threshold, the current detection is determined to have entered a speckle noise-dominated state.

[0037] When the robot approaches a rough metal fence, if the goal remains to reduce bandwidth to improve material outline clarity, speckle will overwhelm useful information, potentially misjudging the fence as a passable gap. This step identifies this inflection point and switches the strategy from maximizing material outline clarity to maximizing the final material identification accuracy. After entering the speckle noise-dominated state, the system appropriately widens the bandwidth within an adjustable range, sacrificing some spectral clarity to reduce speckle and improve overall identification accuracy. The bandwidth operating range is determined from candidate bandwidths that satisfy speckle constraints using the following formula:

[0038] In the formula, B is the candidate spectral line bandwidth, in nm; P(B) is the material determination confidence level under this bandwidth, dimensionless; P max ε represents the highest achievable confidence level among the candidate bandwidths, dimensionless; ε represents the confidence tolerance, dimensionless; ρ(B) represents the noise spectrum ratio under this bandwidth, dimensionless; ρ c The speckle constraint boundary is dimensionless; [B] l B u [] represents the bandwidth operating range, in nm.

[0039] When determining the bandwidth working range, the lower bound of the spectral bandwidth narrowing is determined by constraining the noise spectral ratio to not exceed the preset discrimination threshold; the upper bound of the spectral bandwidth narrowing is determined by constraining the spectral response difference to not be lower than the minimum discrimination required for material classification; the spectral bandwidth narrowing range that maximizes the material determination confidence is searched between the lower and upper bounds, and the corresponding bandwidth range is determined as the bandwidth working range. During the search, each candidate bandwidth can be scanned in a set step between the lower and upper bounds, and its corresponding material determination confidence can be calculated.

[0040] The confidence tolerance ε typically takes a value of 0.05, which can be adjusted within the range of 0.02 to 0.1. Its value is determined based on the calibration results of balancing the confidence concession and the working range width. The speckle constraint limit ρc typically takes a value of 0.8, which can be adjusted between 0.6 and 1.0. Its value is determined based on the calibration results of speckle being within a controllable range after switching. The preset proportional threshold typically takes a value of 0.6, which can be adjusted between 0.4 and 0.8. Its value is determined based on the balance between suppressing occasional false triggers and ensuring timely response.

[0041] Step S6: Adjust the spectral bandwidth of the detection light source through the bandwidth working range and re-execute the detection to obtain the material determination confidence level. Update the collision risk level of the obstacle based on the material determination confidence level and output the robot obstacle avoidance path decision.

[0042] The spectral bandwidth of the detection light source is adjusted according to the bandwidth working range, and the detection is re-executed to obtain the material determination confidence level. Based on this confidence level, the reliability of the material identification result is determined. The identified material category and confidence level are jointly mapped to the collision risk level of the obstacle. For identification results with insufficient confidence, the risk level is increased according to a conservative principle. Based on the updated collision risk level, the path avoidance priority of each obstacle is determined, resulting in the robot's obstacle avoidance path decision. Specifically, the material determination confidence level is obtained by dividing the difference between the spectral response difference obtained from the re-execution detection and the speckle noise amplitude by the spectral response difference.

[0043] In continuous obstacle avoidance in mixed scenarios, the identification needs continuous correction during movement: after bandwidth adjustment, the system returns to the working range for re-detection. If the confidence level recovers, a risk level is assigned based on the identified material. If the confidence level is still insufficient, a higher risk category is applied to avoid collision risk, thus transforming the optical identification results into a motion decision closed loop. The collision risk level is determined jointly by the material category and the confidence level. The material category corresponds to a baseline risk level, and when the confidence level of the material judgment is insufficient, it is conservatively increased by one level and limited. Furthermore, the difference between the current material judgment confidence level and the material judgment confidence level before entering the speckle noise-dominated state is used to obtain the path priority offset. When the path priority offset is negative, the collision risk level of the corresponding obstacle is increased and its path avoidance priority is adjusted, outputting the corrected obstacle avoidance path decision.

[0044] If the confidence level of the material determination is still lower than the set threshold after re-detection within the bandwidth working range, the bandwidth is iteratively fine-tuned and re-detected within the working range. The iteration ends when the confidence level recovers or reaches the upper limit of the iteration. When the iteration still cannot obtain sufficient confidence, the collision risk level of the obstacle is increased and its path avoidance priority is increased according to the obstacle's possible high-risk category, so as to ensure that the robot can still reliably avoid obstacles when the identification fails.

[0045] The collision risk level typically has 4 levels, which can be adjusted between 3 and 6. The value is determined based on the engineering classification of four typical handling methods: negligible, detour, deceleration, and prohibition. The conservative adjustment coefficient typically has 1.0, which can be adjusted between 0.5 and 2.0. The value is determined based on the need to raise the risk level by one level when the confidence level is insufficient. The confidence threshold typically has 0.7, which can be adjusted between 0.6 and 0.85. The value is determined based on the minimum confidence level required for the identification results to enter the decision-making process. The iteration upper limit typically has 3 iterations, which can be adjusted between 2 and 5 iterations. The value is determined based on the balance between convergence and real-time performance.

[0046] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A non-contact photoelectric detection-enabled obstacle avoidance method for robots, characterized in that, include: S1. A detection beam is emitted to the surface of the obstacle by a detection light source with adjustable spectral bandwidth, and the reflected signal returned from the surface of the obstacle is received to obtain spectral response information reflecting the material characteristics of the obstacle. S2. Based on the spectral response information, determine the degree of difference in spectral response that characterizes the distinguishability of different materials; S3. Determine the amplitude of speckle noise, which characterizes signal quality degradation, based on the reflected signal returned from the surface of the obstacle. S4. Calculate the ratio of the speckle noise amplitude to the spectral response difference to obtain the noise spectral ratio. S5. When the noise spectrum ratio exceeds the preset discrimination threshold, it is determined that the current detection has entered the speckle noise-dominated state. The selection criterion for the spectral bandwidth is switched from maximizing the spectral response difference to maximizing the material judgment confidence under the constraint of the noise spectrum ratio, and the bandwidth working range is determined. S6. Adjust the spectral bandwidth of the detection light source through the bandwidth working range and re-execute the detection to obtain the material determination confidence level. Update the collision risk level of the obstacle according to the material determination confidence level and output the robot obstacle avoidance path decision.

2. The method according to claim 1, characterized in that, The detection light source is a narrowband laser emitter. The reflected signal returned from the surface of the obstacle is received by a beam splitting detection array. Each detection unit in the beam splitting detection array has a narrowband filter with a center wavelength arranged at equal intervals at its front end. Each detection unit synchronously collects the intensity value of the reflected light after being transmitted through the corresponding narrowband filter at the same sampling time, and arranges them in order of wavelength channel to form the spectral response information.

3. The method according to claim 1, characterized in that, The step of obtaining the spectral response difference degree characterizing the distinguishability of different materials based on the spectral response information includes: The amplitude deviation of each channel is obtained by subtracting the measured reflected signal intensity value of each wavelength channel from the corresponding intensity value of the pre-stored material reference spectrum. The amplitude deviations of each channel are arranged in the channel sequence to form an amplitude deviation vector; The amplitude deviation vector is normalized to the total energy of the detection light source by taking the Euclidean norm, and the spectral response difference is output.

4. The method according to claim 1, characterized in that, The step of determining the speckle noise amplitude, which characterizes signal quality degradation, based on the reflected signal returned from the obstacle surface includes: Determine the waveform of the reflected signal, and apply a low-pass filter to the waveform of the reflected signal to obtain the baseline envelope signal; Subtract the corresponding baseline envelope signal from the reflected signal waveform to separate the high-frequency intensity fluctuation component; The root mean square value of the high-frequency intensity fluctuation component within the sampling time window is calculated as the speckle noise amplitude of each wavelength channel; The speckle noise amplitude of each wavelength channel is obtained by weighting the average value according to the proportion of the spectral response intensity of each channel to the total response intensity.

5. The method according to claim 1, characterized in that, Also includes: Based on the spatial distribution characteristics of the speckle noise amplitude, the surface roughness gradient of the obstacle is obtained; The roughness speckle coupling strength is evaluated based on the coherence length corresponding to the surface roughness gradient and the current spectral bandwidth. The preset discrimination threshold is adjusted according to the roughness speckle coupling strength, so that the preset discrimination threshold decreases as the roughness speckle coupling strength increases.

6. The method according to claim 5, characterized in that, Determining the surface roughness gradient of the obstacle based on the spatial distribution characteristics of the speckle noise amplitude includes: The amplitude of the speckle noise is read at different receiving angles to form an energy distribution curve of angle versus noise. The energy distribution curves of the angle and noise are fitted with a mixture of Gaussian and exponential functions using least squares to obtain scattering distribution morphology parameters, including the scattering half-width and scattering skewness. Based on the mapping relationship between the half-width at half-maximum of scattering and the root mean square height of surface roughness, and combined with the characterization of the spatial non-uniformity of roughness by scattering skewness, the surface roughness gradient is calculated.

7. The method according to claim 5, characterized in that, The evaluation of roughness speckle coupling strength includes: The dimensionless coupling factor is obtained by calculating the product of the coherence length and the surface roughness gradient. The square of the dimensionless coupling factor is used as the evaluation value of the roughness speckle coupling strength.

8. The method according to claim 1, characterized in that, The determination that the current detection has entered a speckle noise-dominated state includes: The preset discrimination threshold is set as the critical value corresponding to the noise spectrum ratio when the difference between the speckle noise amplitude and the spectral response is equal; the percentage of frames whose noise spectrum ratio exceeds the preset discrimination threshold is counted in multiple consecutive detection cycles; when the percentage of frames exceeds the preset proportion threshold, it is determined that the current detection has entered the speckle noise-dominated state.

9. The method according to claim 1, characterized in that, The determination of the bandwidth working range includes: determining the lower bound of the spectral bandwidth narrowing degree with the noise spectrum ratio not exceeding the preset discrimination threshold as a constraint; determining the upper bound of the spectral bandwidth narrowing degree with the spectral response difference degree not being lower than the minimum discrimination degree required for material classification as a constraint; searching for the spectral bandwidth narrowing degree that maximizes the confidence of material determination between the lower bound and the upper bound, and determining the corresponding bandwidth range as the bandwidth working range.

10. The method according to claim 1, characterized in that, The step of updating the collision risk level of the obstacle based on the confidence level of the material and outputting the robot's obstacle avoidance path decision includes: The difference between the spectral response difference obtained from the re-execution of the test and the speckle noise amplitude is divided by the spectral response difference to obtain the material determination confidence level; The path priority offset is obtained by subtracting the current material determination confidence level from the material determination confidence level before entering the speckle noise-dominated state. When the path priority offset is negative, the collision risk level of the corresponding obstacle is increased and its path avoidance priority is raised, and the corrected obstacle avoidance path decision is output.

Images