Pure position servo mechanical arm based mine telephone answering method without electrical interface

Abstract

The application discloses a pure position servo mechanical arm-based mine telephone answering method without an electrical interface, adopts an edge and local computing power node architecture, offline calibration is performed on a reference audio played after a voice channel is established as a teaching signal, and a fixed pose parameter is determined by taking an acoustic quality parameter as a teaching target; online operation directly moves to capture according to the fixed pose parameter, captures by position pulse feedback confirmation, and then a handset is docked to an acoustic interaction base for acoustic quality standard judgment; standard audio is uploaded to a local area network for identification and matching with a regulation library for response, and after response, logic hang-up verification is performed and the acoustic interaction base is separated; after communication, physical hang-up verification is performed, and the application realizes intelligent response without destroying an intrinsically safe isolation structure.

Description

Technical Field

[0001] This invention relates to the field of mine communication and robotics, specifically to a mine telephone answering method without electrical interface based on a pure position servo robotic arm. Background Technology

[0002] According to the "Coal Mine Safety Regulations," all relevant underground work locations must be equipped with wired dispatch telephones directly connected to the mine dispatch center. The mine dispatch exchange is deployed above ground, and the dispatch telephone lines extend underground to each extension terminal via a safety barrier on the exchange side. The intrinsically safe isolation structure maintained by the safety barrier covers the entire line from the exchange to the underground extension. The regulations establish the legal status of mine dispatch telephones as a core infrastructure for underground safe communication, while imposing stringent compliance constraints on the intelligent transformation of telephone terminals.

[0003] The wired dispatch telephones used in mines do not have digital dialing devices. Call establishment relies on the standard telephone loop current formed in the line after the terminal goes off-hook. The dispatch exchange identifies the terminal's call status by detecting the state transition between closed and open loops: when the handset is off-hook, the line is open and the loop current is approximately zero; when the handset is picked up, the line is closed, forming a DC loop. The exchange detects the loop current, determines that the handset is off-hook, and connects the call path; when the handset is put back, the line returns to an open loop, the exchange determines that the handset is off-hook, and releases the call path.

[0004] The regulations explicitly prohibit dispatch telephones from being powered locally underground and from being connected to dispatch exchanges via active repeaters. If a conventional intelligent upgrade approach is adopted—that is, by breaking wires in the line to connect an AI voice gateway, electrical signal splitter, or series active repeater equipment—three fundamental obstacles will be encountered: First, the introduction of active equipment directly violates the regulations' prohibitions on local underground power supply and active repeaters; second, the introduction of additional electrical components will change the DC loop parameters of the line, causing the exchange to be unable to detect the standard loop current signal and misjudge the terminal status; third, it may damage the intrinsically safe isolation structure maintained by the safety barrier, causing the communication circuit to lose its explosion-proof compliance. Therefore, any solution relying on direct electrical access carries significant hidden dangers.

[0005] Even disregarding compliance constraints, the acoustic coupling of the handset to the mechanical actuator presents significant obstacles. As a passive acoustic device, the handset's acoustic radiation is directional, and the opening shape, radiation surface curvature, and sound outlet distribution vary depending on the handset model. If there is a positional offset, angular deflection, or gap between the pickup device and the handset, the acoustic coupling quality will significantly decrease, resulting in a signal-to-noise ratio and speech intelligibility index below the identifiable threshold, causing subsequent speech recognition to fail. The accuracy of visual pose calculation is limited by the resolution of the image acquisition device and algorithm errors. Relying solely on visual calculation for a single clamping operation cannot guarantee optimal acoustic coupling; a closed-loop feedback mechanism between acoustic coupling quality and pose adjustment is needed. Furthermore, dispatch telephones lack open electrical interfaces, and the disconnection status cannot be directly obtained via electrical signals. Relying solely on disconnection tone verification is prone to misjudgment due to differences in handset models and fluctuations in line status, potentially leading to handset deadlock or busy lines.

[0006] Without altering existing communication lines, compromising inherently safe isolation structures, or adding active equipment underground, employing mechanical actuators to simulate manual hook-up / hook-up actions is a technical approach that complies with coal mine safety regulations. However, this approach still faces the following unresolved technical challenges: First, how to reliably confirm the earpiece's gripping status without torque sensor feedback; second, how to achieve high-quality acoustic coupling between the pickup device and the earpiece receiver through fine-tuning the robotic arm's end-effector pose without electrical interface constraints; third, how to reliably confirm the hook-up status without electrical feedback signals; and fourth, how to perform high-performance voice recognition and semantic responses in a physically isolated environment where connection to the external public internet is strictly prohibited.

[0007] Existing solutions that simulate human operation of physical interfaces through mechanical devices (such as simulating human operation of steering wheels and pedals in the field of unmanned dump trucks for mining) are usually open-loop executions, without establishing a closed-loop feedback optimization mechanism for the physical characteristics of the interactive object. The teach-reproduce paradigm in the field of industrial robots also does not involve scenarios where acoustic coupling quality parameters are used as teaching targets. The stringent physical constraints faced by this application are as follows: First, acoustic signals exhibit reverberation, standing waves, and non-monotonic responses under near-field coupling, and a single index (such as signal-to-noise ratio) cannot stably characterize the coupling quality. A joint judgment logic of both signal-to-noise ratio and speech intelligibility index is required to solve the problem of determining directional effectiveness. Second, the lack of electrical interface constraints makes it impossible to obtain electrical feedback such as loop current as auxiliary observation. Visual feedback can only obtain spatial geometric pose information and cannot directly characterize the communication coupling quality. Acoustic feedback becomes the only observation method that can directly reflect the communication effect. This constraint forces the specific selection of optimization strategies and degradation mechanisms. Third, the physical coupling optimization problem between the end effector of the robotic arm and the passive acoustic device has not been revealed in existing literature. Therefore, the aforementioned conventional technical means present non-obvious difficulties in combining with the stringent physical constraints faced by this application. Summary of the Invention

[0008] (a) Technical issues

[0009] This invention provides a method for answering mine dispatch telephones without electrical interfaces based on a pure position servo robotic arm. It aims to solve the technical challenge of existing mine dispatch telephones being unable to directly interface with intelligent software under intrinsically safe isolation constraints and the lack of electrical interfaces. Specifically: First, how to confirm the gripping state of the handset by the robotic arm's end-effector without torque sensor feedback; Second, how to achieve stable acoustic coupling between the handset's receiving and transmitting ends and the pickup device and speaker via an independent acoustic interaction base, and how to achieve optimal coupling immediately upon receiving a call, avoiding response delays caused by successive optimizations, all under the condition of no electrical feedback signal; Third, how to reliably confirm the on-hook status to avoid phone deadlock and busy signals, under the condition of no electrical feedback signal; Fourth, how to complete high-performance voice recognition and semantic response tasks in a physically isolated environment where connection to the external public internet is strictly prohibited.

[0010] (II) Technical Content

[0011] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: a mine telephone answering method without electrical interface based on a pure position servo robotic arm, employing a distributed communication architecture including edge computing nodes of the robotic arm and physically isolated local computing nodes in the mining area. An acoustic interaction base is fixedly installed within the working range of the robotic arm. The acoustic interaction base is provided with a first groove structure and a second groove structure. A sound pickup device is installed on the inner wall of the first groove structure, and a miniature speaker is installed on the inner wall of the second groove structure. The method includes the following steps executed in response to an incoming call event:

[0012] Ring detection steps: The edge computing node acquires an image of the phone's ring indicator light through an image acquisition device and acquires an ambient audio signal through an ambient microphone. It performs ringing spectrum feature recognition on the ambient audio signal and ring indicator light status recognition on the image. When both determine that a ringing event exists, it identifies the phone from which the ringing occurs and triggers the off-hook process.

[0013] Off-hook and gripping steps: The edge computing node selects the corresponding pre-determined fixed posture parameters according to the phone from which the ringing is coming from, controls the robotic arm to move to the handset position according to the fixed posture parameters, controls the end gripping part to perform a closing action, and determines the gripping state based on the position pulse feedback value of the gripping part servo motor;

[0014] Acoustic docking steps: After successful grabbing, control the robotic arm to align the receiver and transmitter of the earpiece and insert them into the first groove structure and the second groove structure respectively;

[0015] Acoustic quality judgment steps: The edge computing node acquires the call audio signal collected by the pickup device, extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters, and judges whether the acoustic quality parameters are lower than the preset quality threshold.

[0016] Voice recognition and response steps: If the acoustic quality parameters are not lower than the quality threshold, the call audio signal is transmitted to the physically isolated local computing node via the mine's internal LAN for localized voice recognition and voiceprint recognition to obtain the call text content and caller identity information. The call text content and caller identity information are matched with the pre-stored coal mine safety regulations knowledge base to generate response audio data, which is then sent to the edge computing node via the mine's internal LAN and played through a miniature speaker.

[0017] Hang-up verification steps: After the response playback is completed, the second-level logical hang-up verification is triggered to detect the call termination status. The robotic arm is controlled to remove the earpiece from the first groove structure and the second groove structure. Then, the first-level physical hang-up verification is triggered to confirm that the earpiece is physically back in place. An event chain record is generated based on the results of the second-level logical hang-up verification and the first-level physical hang-up verification.

[0018] Furthermore, the gripping state is determined based on the position pulse feedback value of the servo motor of the gripping part. Specifically: if the position pulse feedback value is less than or equal to the preset no-load closing threshold, the gripping is determined to have failed; if the position pulse feedback value is within the preset effective gripping range, the gripping is determined to have succeeded; if the position pulse feedback value is greater than the no-load closing threshold and less than the lower limit of the effective gripping range, or greater than the upper limit of the effective gripping range and less than the lower limit of the complete release tolerance range, the gripping state is determined to be abnormal. When the gripping fails or the gripping state is abnormal, the robotic arm is controlled to reset and re-execute the disengagement and gripping steps until the gripping is successful or the number of retries reaches the preset retry limit, after which an abnormal alarm is triggered.

[0019] The preset no-load closure threshold is the position pulse feedback value of the servo motor that deviates from the fully closed position of the clamping part within a preset tolerance range. The preset effective clamping interval is a continuous interval within a preset tolerance range centered on the average position pulse feedback value of the servo motor when the outer diameter of the earpiece is stably clamped.

[0020] Furthermore, the quality threshold includes a preset signal-to-noise ratio threshold and a preset clarity threshold. In the acoustic quality judgment step, if the signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold or the speech clarity index is lower than the preset clarity threshold, the subsequent automatic recognition process is stopped, the current communication link is transferred to the manual dispatch terminal, an event chain record carrying a manual review mark is generated, and an event mark suggesting recalibration is generated.

[0021] After being transferred to the manual dispatch terminal, the edge computing node controls the robotic arm to maintain the clamping state and remove the earpiece from the first groove structure and the second groove structure, and move to the preset handover posture to wait for the manual removal of the earpiece; during the waiting period, the position pulse feedback value of the servo motor of the clamping part is read cyclically. If the position pulse feedback value drops below the no-load closing threshold, it is determined that the earpiece has been removed, and the clamping part is controlled to open and retreat to the standby posture.

[0022] Furthermore, after generating the event flag suggesting recalibration, the edge computing node identifies the calibrable time window based on historical call frequency statistics and provides a prompt. After manual confirmation by the dispatcher, the operator manually initiates a call to the phone to be calibrated from another phone to trigger the offline calibration step; or the offline calibration step is executed in response to the dispatcher's active triggering instruction.

[0023] Furthermore, the fixed pose parameters are predetermined through the following offline calibration steps:

[0024] Control the robotic arm to move to the earpiece position for visual pre-alignment, and record the current pose as baseline coordinates;

[0025] The call is initiated manually from another phone to the phone to be calibrated. After the phone rings, the clamping part grabs the handset and the handset is picked up to establish a call path. The robotic arm aligns the receiving end and the transmitting end of the handset with the first groove structure and the second groove structure respectively. The calling phone plays the pre-recorded reference audio.

[0026] Activate the coordinate descent heuristic pose optimization strategy based on directional continuation, and sequentially control the robotic arm to perform step translation along the preset axis of the end-effector coordinate system or deflect at a fixed angle around the preset axis. After each fine adjustment, the acoustic quality parameters are re-extracted. When the signal-to-noise ratio and speech intelligibility index do not decrease after a single fine adjustment, and the increase of at least one indicator is greater than the preset fluctuation threshold, the fine adjustment direction is confirmed to be effective and retained, and the iteration continues until the signal-to-noise ratio is not lower than the preset signal-to-noise ratio threshold and the speech intelligibility index is not lower than the preset intelligibility threshold, or the number of iterations reaches the preset upper limit.

[0027] If the acoustic quality parameters are still lower than the preset quality threshold when the preset upper limit is reached, the pose data with the best acoustic quality parameters in each iteration is called as the candidate optimal pose. The acoustic quality parameters corresponding to the candidate optimal pose and the baseline coordinates are compared. If the acoustic quality parameters of the candidate optimal pose are lower than the acoustic quality parameters corresponding to the baseline coordinates, the baseline coordinates are restored and the calibration is marked as failed. Otherwise, the candidate optimal pose is locked to the fixed pose parameters corresponding to the earpiece and stored persistently.

[0028] The above calibration process was performed independently on the two telephones on both sides of the dispatch console, and the fixed posture parameters were recorded respectively.

[0029] Furthermore, after the local computing node performs localized speech recognition and voiceprint recognition to obtain the call text content and caller identity information, it also includes a three-factor identity joint verification: the local computing node obtains the caller extension number and the corresponding fixed work location information, obtains the current shift schedule data stored locally, and performs a consistency comparison of the caller identity information, the caller extension number, and the shift schedule data. Only when the three match completely will the automatic response permission involving security commands be granted; otherwise, the communication link will be transferred to the manual dispatch console.

[0030] Furthermore, before the edge computing node extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters, it also includes: using an adaptive filter to suppress background noise in the call audio signal, wherein the reference noise input of the adaptive filter is the background noise signal of the dispatch room environment collected by an environmental microphone independently set outside the mine dispatch telephone.

[0031] Furthermore, the local area network within the mining area is a dedicated network that is connected to the mining area's intranet access point through wireless communication modules deployed on edge computing nodes, or directly connected to the network cable through a switch, and there is no data routing or exchange between the physically isolated local computing nodes and the external public Internet.

[0032] Furthermore, both the first groove structure and the second groove structure are funnel-shaped guide structures that gradually expand in the opening direction. They are used to guide the receiver and transmitter of the earpiece into the predetermined position of the groove through the inclined surface of the groove when there is a deviation in the docking posture of the robotic arm.

[0033] Furthermore, before controlling the robotic arm to move to the earpiece position according to the fixed pose parameters, the edge computing node also compares the calculated current pose of the earpiece with the fixed pose parameters to confirm that the earpiece is still in the expected position.

[0034] Furthermore, in the hang-up verification step, the second-level logical hang-up verification is as follows: within a preset time window after the response playback is completed, the microphone continuously monitors to detect the busy tone characteristics of the communication network and the human voice characteristics; the first-level physical hang-up verification is as follows: the image acquisition device confirms that the handset has entered the preset return area of ​​the handset base, and detects that the position pulse feedback value of the clamping part servo motor has reached the preset release tolerance range.

[0035] Furthermore, an event chain record is generated based on the results of the second-level logical hang-up verification and the first-level physical hang-up verification. Specifically: if the first-level physical hang-up verification passes and the conditions of the second-level logical hang-up verification are met, a normal closed-loop event chain record is generated; if the first-level physical hang-up verification passes but the conditions of the second-level logical hang-up verification are not met, an event chain record with an exception flag indicating that it needs to be reviewed is generated; if the first-level physical hang-up verification fails, the hang-up action is re-executed until the first-level physical hang-up verification passes or the number of retries reaches the preset limit.

[0036] (III) Technical Effects

[0037] The advantages of this invention compared to the prior art are:

[0038] 1. To address the issue of hang-up confirmation without electrical feedback, this invention employs a multimodal cascaded verification mechanism. A second-level logical hang-up verification is executed first, consisting of busy tone feature extraction and human voice timeout detection. A first-level physical hang-up verification is executed subsequently, consisting of visual repositioning confirmation and position pulse feedback. A degradation processing path based on the physical level results is also established. This mechanism achieves hang-up confirmation without electrical feedback at both logical and physical levels, avoiding the deadlock / busyline problem caused by relying solely on disconnection tone verification under conditions without electrical feedback.

[0039] 2. To address the issue of gripping confirmation without torque sensors, this invention extracts the position pulse feedback value of the servo motor of the clamping part and sets an unloaded closing threshold and an effective clamping range. This achieves reliable non-tactile determination of the earpiece gripping state using a pure position servo method. This solution avoids the application limitations of torque sensors under intrinsically safe isolation constraints, and the determination is based on the inherent position feedback of the servo motor, requiring no additional electrical components and meeting the intrinsic safety requirements for mining applications.

[0040] 3. To address the physical alignment problem of passive acoustic devices without electrical interface constraints, this invention adopts a scheme combining an independent acoustic interaction base with a two-stage architecture that separates offline calibration and online operation. An acoustic interaction base is fixedly installed within the working range of the robotic arm, with two groove structures on it to respectively accommodate the receiver and transmitter of the earpiece. A pickup device and a miniature speaker are respectively installed on the inner walls of the grooves. In the offline calibration stage, acoustic coupling quality parameters are used as the teaching target, and the reference audio played after the voice path is established is used as the teaching signal. This teaching signal and the input signal used for acoustic quality parameter extraction in the online operation stage are of the same type of reference speech signal, ensuring that the coupling quality optimized in the calibration stage is consistent with the coupling quality determined in the online operation stage in terms of signal characteristics. During the online operation phase, the earpiece is directly inserted into the corresponding groove of the acoustic interaction base based on the fixed pose parameters determined in the offline calibration phase. This avoids the response delay caused by successive optimization and ensures that the optimal acoustic coupling state determined by the offline calibration phase is reached immediately upon receiving a call. The offline calibration phase uses AND dual-threshold convergence logic for two acoustic indicators, signal-to-noise ratio and speech intelligibility index, to solve the problem of determining the directional validity under non-monotonic acoustic mapping. A coordinate restoration protection mechanism is also set to prevent the writing of pose parameters that are inferior to the initial visual pre-alignment results.

[0041] 4. To address the high-computing-power processing challenges in physically isolated environments, this invention employs an asymmetric computing power allocation architecture between edge computing nodes and physically isolated local computing nodes within the mining area. Edge nodes handle low-computing-power real-time control tasks, including pose calculation, position pulse reading, and signal-to-noise ratio calculation, and perform pose optimization during offline calibration. Local computing nodes handle high-computing-power business logic, including speech recognition, voiceprint recognition, procedure library matching, and response audio generation. Both communicate via the mining area's internal local area network, and there is no data routing or exchange between the physically isolated local computing nodes and the external public internet. This architecture meets the data security isolation requirements of the mining area, physically preventing data leakage risks and complying with national data privacy and safety production regulations.

[0042] 5. This invention is a system-level solution proposed to address the stringent technical constraints of mining areas lacking electrical interfaces and requiring network isolation. It combines pure vision and acoustic degradation cascading for on-the-go confirmation, breaking the deadlock problem under conditions without electrical feedback. At the edge, it uses gripper servo position pulse feedback to replace torque sensors to achieve non-tactile grasping confirmation. It also overcomes the blind grasping and sound leakage problem caused by insufficient physical alignment accuracy of passive acoustic devices by using a two-stage architecture that determines fixed pose parameters offline and directly reuses them online. The architecture constructs an asymmetric collaborative mode of real-time edge physical control and local high-computing-power semantic processing. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the distributed topology of the system for the mine telephone answering method based on a pure position servo robotic arm without electrical interface according to the present invention.

[0044] Figure 2 This is a schematic diagram of the acoustic interactive base of the present invention.

[0045] Figure 3 This is the main control flowchart of the mine telephone answering method of the present invention.

[0046] Figure 4 This is a flowchart of the offline calibration phase of the present invention.

[0047] Figure 5 This is a flowchart of the online operation phase of the present invention.

[0048] Reference numerals: 101, End-effector node; 102, Edge computing node; 103, Local area network within the mine; 104, Physically isolated local computing node; 105, Mine dispatch telephone; 116, Telephone stand; 117, Ringing indicator light; 106, Safety barrier; 107, Underground extension terminal; 108, Acoustic interaction base; 112, Image acquisition device; 113, End-effector clamping part; 114, Clamping part servo motor; 120, Preset reference mark; 118, First groove structure; 119, Second groove structure; 109, Sound pickup device; 110, Miniature speaker; 115, Handpiece; 111, Ambient microphone. Detailed Implementation

[0049] The present invention will now be described in further detail with reference to the accompanying drawings.

[0050] Combined with appendix Figure 1 To be continued Figure 5A method for answering mine dispatch telephone calls without electrical interfaces based on a pure position servo robotic arm is disclosed. The method employs a distributed communication architecture including edge computing nodes of the robotic arm and physically isolated local computing nodes within the mining area. An acoustic interaction base 108 is fixedly installed within the working range of the robotic arm. The acoustic interaction base 108 has a first groove structure 118 and a second groove structure 119. A sound pickup device 109 is installed on the inner wall of the first groove structure 118, and a miniature speaker 110 is installed on the inner wall of the second groove structure 119. The method includes the following steps executed sequentially:

[0051] S1: The edge computing node 102 acquires an image of the mine dispatch telephone 105 through the image acquisition device 112 set at the end of the robotic arm, extracts the visual features of the preset reference mark 120, and calculates the current pose of the handset 115 relative to the coordinate system of the end tool of the robotic arm.

[0052] S11: The edge computing node 102 collects ambient audio signals through an environmental microphone 111 independently set outside the mine dispatch telephone, and simultaneously acquires an image of the telephone ringing indicator light 117 through the image acquisition device 112. It performs ringing spectrum feature recognition on the ambient audio signal and ringing indicator light status recognition on the image. When both determine that a ringing event exists, it identifies the position of the ringing indicator light 117 in the image to determine the telephone from which the ringing originated, and triggers the subsequent off-hook process.

[0053] S2: The edge computing node 102 selects the corresponding pre-determined fixed pose parameters according to the ringing source telephone determined in S11, compares the current pose calculated in S1 with the fixed pose parameters to confirm that the handset 115 is still in the expected position, controls the robotic arm to move directly to the corresponding handset 115 position according to the fixed pose parameters, and controls the end gripper 113 to perform a closing action; extracts the position pulse feedback value of the gripper servo motor 114, and if the position pulse feedback value is less than or equal to the preset no-load closing threshold, it determines that the gripper is in a certain position. If the grasping fails, the robotic arm is reset and step S1 is re-executed until the grasping is successful or the number of retries reaches the preset retry limit, triggering an abnormal alarm. If the pulse feedback value at this position is within the preset effective gripping range, the grasping is considered successful. If the pulse feedback value at this position is greater than the no-load closing threshold and less than the lower limit of the effective gripping range, or greater than the upper limit of the effective gripping range and less than the lower limit of the complete release tolerance range, the gripping state is considered abnormal, the robotic arm is reset and step S1 is re-executed until the grasping is successful or the number of retries reaches the preset retry limit, triggering an abnormal alarm.

[0054] S22: Control the robotic arm to align the receiving end and transmitting end of the handset 115 with the fixed posture parameters and insert them into the first groove structure 118 and the second groove structure 119 respectively.

[0055] S3: The edge computing node 102 acquires the call audio signal collected by the pickup device 109, and extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters.

[0056] S4: The edge computing node 102 determines whether the acoustic quality parameter is lower than a preset quality threshold; if the signal-to-noise ratio is not lower than the preset signal-to-noise ratio threshold and the speech clarity index is not lower than the preset clarity threshold, then step S5 is executed; if the signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold or the speech clarity index is lower than the preset clarity threshold, then the subsequent automatic recognition process is stopped, the current communication link is transferred to the manual dispatch terminal, an event chain record carrying a manual review mark is generated, and an event mark suggesting recalibration is generated; wherein, the fixed pose parameter is predetermined through an offline calibration stage, the offline calibration stage (the offline calibration referred to in this application refers to the parameter calibration stage performed before the formal operation of the system or during idle periods when there is no response to incoming calls, and is still in the local area network connection state within the mining area, rather than operating outside the network) executes a coordinate descent heuristic pose optimization strategy based on directional continuation during idle periods when there is no response to incoming calls to determine the fixed pose parameter and persist it;

[0057] S5: The edge computing node 102 transmits the call audio signal whose acoustic quality parameters have reached the quality threshold to the physically isolated local computing node 104 via the local area network 103 in the mining area; the physically isolated local computing node 104 performs localized speech recognition and voiceprint recognition to obtain the call text content and caller identity information.

[0058] S6: The physically isolated local computing node 104 matches the call text content and caller identity information with the pre-stored coal mine safety regulations knowledge base, generates response audio data, and sends the response audio data to the edge computing node 102 via the mine's internal local area network 103, and plays it through the miniature speaker 110.

[0059] S66: After the response playback is completed, the second level of logic hang-up verification is triggered: within the preset time window after the response playback is completed, the pickup device 109 continuously monitors to detect the busy tone characteristics of the communication network and the human voice characteristics. After the preset time window expires, the robotic arm is controlled to maintain the clamping state and remove the earpiece 115 from the first groove structure 118 and the second groove structure 119.

[0060] S7: The edge computing node 102 controls the robotic arm to move the handset to the top of the handset holder 116 according to a preset trajectory, and controls the end gripper 113 to perform a release action, triggering the first-level physical hang-up verification: the image acquisition device 112 confirms that the handset 115 has entered the preset return area of ​​the handset holder 116, and detects that the position pulse feedback value of the gripper servo motor 114 has reached the preset release tolerance range, then the physical hang-up action is determined to be completed; if the physical hang-up action is completed and the previous second-level logical hang-up verification conditions have been met, then a normal closed-loop event chain record is generated; if the physical hang-up action is completed but the previous second-level logical hang-up verification conditions have not been met, then an event chain record carrying an abnormality mark to be reviewed is generated; if the physical hang-up action is not completed, then the hang-up action is re-executed until the physical hang-up action is completed or the number of retries reaches the preset upper limit.

[0061] In this embodiment, the preset no-load closure threshold is the position pulse feedback value of the servo motor that deviates from the fully closed position of the end clamping part 113 within a preset tolerance range, and the preset effective clamping interval is a continuous interval within a preset tolerance range centered on the average position pulse feedback value of the servo motor when the outer diameter of the earpiece 115 is stably clamped.

[0062] In this embodiment, the fixed pose parameters are predetermined through the following offline calibration steps: The robotic arm is controlled to move to the position of the handset 115 for visual pre-alignment, and the current pose is recorded as baseline coordinates; a call is manually initiated from another phone to the phone to be calibrated, and after ringing, the end effector 113 is controlled to grasp the handset 115 and establish a call path; the robotic arm is controlled to align the receiving end and the transmitting end of the handset 115 and insert them into the first groove structure 118 and the second groove structure 119 respectively; the calling phone plays a pre-recorded reference audio, which serves as the input signal for extracting the signal-to-noise ratio and speech clarity index of the call audio signal; a coordinate descent heuristic pose optimization strategy based on directional continuation is activated, and the robotic arm is sequentially controlled to perform step translation along the preset axis of the end effector coordinate system or to perform a fixed-length angle deflection around the preset axis; after each fine adjustment, the pose arrival is confirmed by the image acquisition device, and the position is re-evaluated. Acoustic quality parameters are taken. When the signal-to-noise ratio (SNR) and speech intelligibility index do not decrease after a single fine-tuning, and the increase in at least one indicator is greater than a preset fluctuation threshold, the fine-tuning direction is confirmed as valid and retained, and iteration continues until the SNR is not lower than a preset SNR threshold and the speech intelligibility index is not lower than a preset intelligibility threshold, or the number of iterations reaches a preset upper limit. If the acoustic quality parameters are still lower than the quality threshold when the preset upper limit is reached, the pose data with the best acoustic quality parameters in each iteration is called as the candidate optimal pose. The acoustic quality parameters corresponding to the candidate optimal pose and the baseline coordinates are compared. If the acoustic quality parameters of the candidate optimal pose are lower than the acoustic quality parameters corresponding to the baseline coordinates, the baseline coordinates are restored and the calibration is marked as failed. Otherwise, the candidate optimal pose is locked as the fixed pose parameters corresponding to the handset and stored persistently. The above calibration process is performed independently on the handsets on both sides of the dispatch console, and the fixed pose parameters are recorded respectively.

[0063] In this embodiment, the coordinate descent heuristic pose optimization strategy based on directional continuation has an initial step translation displacement of 10mm to 30mm, preferably 20mm, and a minimum step translation displacement of 10mm; the single deflection of the fixed-length angle deflection is 2 degrees to 8 degrees, preferably 5 degrees; confirming that the fine-tuning direction is valid and retaining it includes: when a certain direction is determined to be valid, continuing to step along that direction until it becomes invalid; if no valid direction can be found in any of the positive and negative directions of all axes, the step size is halved and the search is restarted.

[0064] In this embodiment, after generating the event flag for suggested recalibration, the edge computing node 102 identifies the calibrable time window based on historical call frequency statistics and provides a prompt. After manual confirmation by the dispatcher, the operator manually initiates a call to the phone to be calibrated from another phone to trigger the offline calibration step; or the offline calibration step is executed in response to the dispatcher's active triggering instruction.

[0065] In this embodiment, after the physically isolated local computing node 104 performs localized voice recognition and voiceprint recognition to obtain the call text content and caller identity information, it also includes a three-factor identity joint verification: the physically isolated local computing node 104 obtains the caller extension number and the corresponding fixed work location information, obtains the local stored current shift schedule data, and performs a consistency comparison of the caller identity information, the caller extension number and the shift schedule data. Only when the three are completely matched will the automatic response permission involving security commands be opened; otherwise, the communication link will be transferred to the manual dispatch console.

[0066] In this embodiment, before the edge computing node 102 extracts the signal-to-noise ratio and speech clarity index of the call audio signal as acoustic quality parameters, it further includes: using an adaptive filter to suppress background noise in the call audio signal. The reference noise input of the adaptive filter is the background noise signal of the dispatch room environment collected by the environmental microphone 111 independently set outside the mine dispatch telephone 105.

[0067] In this embodiment, the local area network inside the mining area is a dedicated network that is connected to the access point of the mining area network through a wireless communication module deployed on the edge computing node 102, or directly connected to the network cable through a switch, and there is no data routing exchange between the physically isolated local computing node 104 and the external public Internet.

[0068] In this embodiment, both the first groove structure 118 and the second groove structure 119 are funnel-shaped guide structures that gradually expand in the opening direction. When there is a deviation in the docking posture of the robotic arm, the receiving end and the transmitting end of the earpiece are guided into the predetermined position of the groove through the inclined surface of the groove, providing auxiliary positioning accuracy for mechanical self-centering and reducing the dependence on pure vision servo positioning accuracy.

[0069] The present invention will be further described in detail below through a specific implementation process.

[0070] All data transfers involving sensitive information such as caller voiceprints and work schedules are completed via the M5StackBasic embedded edge device through the mine's private LAN to a private server node in the local data center, without calling any artificial intelligence interfaces on the external public internet. This purely localized, physically isolated architecture meets the data security isolation requirements of the mining area, physically preventing the risk of data leakage and complying with national data privacy and safety production regulations.

[0071] The system physical architecture used in this embodiment is as follows: Figure 1 As shown, it consists of four parts: end execution node 101, edge computing node 102, local area network within the mining area 103, and physically isolated local computing power node 104.

[0072] The end effector 101 employs a myCobot 280 M5Stack six-axis collaborative robotic arm with an effective working radius of 280mm, a rated load of 250g, and a repeatability of ±0.5mm, meeting the gripping load requirements of the handset 115 (typically weighing approximately 150 to 200g) of the mine dispatch telephone 105. The mine dispatch telephone 105 is an intrinsically safe wired dispatch telephone for mining. Its body is equipped with a handset base 116 and a ringing indicator light 117. The handset base 116 supports the handset 115 and includes a built-in fork spring hang-up switch. When the handset 115 is placed on the handset base 116, the fork spring is pressed down, resulting in an on-hook state. The ringing indicator light 117 is located on the front of the telephone body and illuminates or flashes when a call rings. A preset reference mark 120 (ArUco code in this embodiment) is affixed to the surface of the telephone body as a manual reference point for visual pose calculation. The robotic arm employs a dual-core architecture: the base houses an M5Stack Basic main controller (using an ESP32-D0WDQ6-V3 dual-core processor with a clock speed of 240MHz) acting as edge computing node 102; the end effector houses an M5Stack ATOM secondary controller (using an ESP32 processor) acting as the core for end effector data aggregation and servo control coordination. Communication between the base main controller and the end effector is via serial port at a baud rate of 115200bps, with control latency in the millisecond range.

[0073] The edge computing node 102 is deployed on the robotic arm base and is responsible for low-computing-power real-time tasks, including reading the position pulse feedback value of the servo motor 114 of the gripping part, solving the visual features of the preset reference mark 120, calculating the signal-to-noise ratio and voice clarity index of the call audio, and executing the pose optimization control strategy under acoustic quality feedback.

[0074] The physically isolated local computing node 104 is deployed in the surface dispatch room or surface computer room and is a private server. This server communicates with the edge computing node 102 via the mine's internal LAN 103. Communication is achieved through a wireless communication module deployed on the edge computing node 102 connecting to the mine's internal wireless access point, or through a direct network cable connection via a switch. There is no data routing or exchange between the physically isolated local computing node 104 and the external public internet. This server is responsible for high-computing-power business logic, including receiving compliant audio from the LAN, running locally deployed speech recognition, voiceprint recognition, and coal mine safety regulations knowledge base comparison, generating response audio data, and then sending it back to the edge terminal via the LAN.

[0075] like Figure 2 As shown, the acoustic interaction structure in this embodiment is arranged according to the following spatial relationship:

[0076] An image acquisition device 112 is fixedly installed 5cm in front of the end flange of the robotic arm to acquire images of the mine dispatch telephone 105 and the status images of the ringing indicator light 117. A left and right gripping finger are symmetrically arranged below the end flange, forming an adaptive parallel gripper. This gripper is driven by a servo motor 114 and controlled by position pulses. The protocol-level control range of the position pulses is 0 to 100, covering the complete movement stroke of the gripper from fully open to fully closed. Lower pulse values ​​correspond to the closed state, and higher pulse values ​​correspond to the open state. The end gripping part 113 is a standard adaptive parallel gripper used to grip the holding section (waist) of the handset 115, without additional fixed pickup devices or speakers.

[0077] An acoustic interaction base 108 is fixedly installed within the working range of the robotic arm (for example, on a fixed bracket next to the base of the robotic arm or near the mine dispatch telephone 105). The acoustic interaction base 108 is provided with a first groove structure 118 and a second groove structure 119. A sound pickup device 109 is provided on the inner wall of the first groove structure 118, with the sound pickup surface facing the groove opening direction; a miniature speaker 110 is provided on the inner wall of the second groove structure 119, with the sound emitting surface facing the groove opening direction. The distance between the first groove structure 118 and the second groove structure 119 is preset according to the actual length of the handset 115 of the mine dispatch telephone 105 to be adapted. Both the first groove structure 118 and the second groove structure 119 are horn-shaped guide structures that gradually expand towards the opening direction. When there is a deviation in the docking posture of the robotic arm, the receiving end and the transmitting end of the handset 115 are guided into the predetermined position of the groove through the groove slope, providing auxiliary positioning accuracy for mechanical self-centering and reducing the dependence on pure visual servo positioning accuracy.

[0078] An ambient microphone 111 is independently installed next to the phone stand 116 of the mine dispatch telephone 105. The ambient microphone 111 is fixed on an independent bracket and does not contact the telephone body to avoid vibration coupling. It is only used to collect the background noise of the dispatch room environment as a reference signal. The target signal collected by the pickup device 109 in the first groove structure 118 of the acoustic interaction base 108 is completely independent of the reference signal collected by the ambient microphone 111. The former is used for voice extraction, and the latter is used as the reference noise input for the adaptive filter.

[0079] Before executing step S1, the edge computing node 102 performs system initialization and self-test procedures. First, the edge computing node 102 issues a self-test command for the robotic arm to confirm that the robotic arm is not in a curled or joint collision state. Then, the base main controller detects the connection status of the end-effector through the handshake signal detection interface. When the handshake is successful, it is determined that the end-effector connection is normal. When the handshake fails or the connection times out, a hardware alarm is triggered and the program is terminated to prevent the robotic arm from going out of control in an unlocked state.

[0080] The edge computing node 102 acquires images of the mine dispatch telephone 105 through the image acquisition device 112 set at the end of the robotic arm, and extracts the visual features of the preset reference mark 120 (ArUco code in this embodiment) pre-attached to the telephone body. Since the handset 115 is stationary on the telephone stand 116 when the phone is ringing, the relative positional relationship between the telephone stand 116, the telephone body, and the reference mark 120 is fixed and known. Therefore, after the visual features of the reference mark 120 are extracted by the image acquisition device 112 to calculate the telephone body pose, the current pose of the handset 115 relative to the tool coordinate system at the end of the robotic arm is indirectly obtained through known coordinate transformation. The current pose is used to compare with the predetermined fixed pose parameters to confirm that the handset 115 is still in the expected position.

[0081] In response to the limitations of the image acquisition device 112 resolution due to changes in lighting in the control room, the image acquisition device 112, in conjunction with a preset reference marker 120, acquires contour features to calculate the current pose, thus avoiding interference from ambient lighting fluctuations on the accuracy of visual calculation.

[0082] The edge computing node 102 collects ambient audio signals through an environmental microphone 111 independently located next to the phone stand 116, and simultaneously acquires an image of the phone ringing indicator light 117 through an image acquisition device 112. It performs ringing spectrum feature recognition on the ambient audio signals and ringing indicator light status recognition on the image. When both determine that a ringing event exists, it further identifies the position of the ringing indicator light 117 in the image to determine whether the phone originating from the ringing is the phone on the left or right side of the dispatch console, and triggers the subsequent off-hook process. This dual-mode joint detection mechanism avoids false triggering under environmental interference or light fluctuation conditions with a single detection method. The determination result of the phone originating from the ringing is used in the subsequent step S2 to select the corresponding fixed pose parameters to support independent calibration and differentiated reuse of phones on both sides of the dispatch console.

[0083] The edge computing node 102 selects the corresponding pre-determined fixed pose parameters based on the ringing source phone determined in S11, compares the current pose calculated in S1 with the fixed pose parameters to confirm that the handset 115 is still in the expected position, controls the robotic arm to move directly to the corresponding handset position according to the fixed pose parameters, and controls the end gripper 113 to perform a closing action. Then, it extracts the position pulse feedback value of the gripper servo motor 114 and confirms the gripping state according to the following determination logic:

[0084] When the position pulse feedback value is less than or equal to the preset no-load closing threshold, it is determined that the gripper has closed completely without contacting the earpiece, and the grasping is determined to be a failure. The robotic arm is controlled to reset and the visual pose calculation of step S1 is re-executed until the grasping is successful or the number of retries reaches the preset retry limit, and then an abnormal alarm is triggered.

[0085] When the position pulse feedback value is within the preset effective gripping range, it is determined that the gripper has stably gripped the earpiece, and the gripping is considered successful.

[0086] If the position pulse feedback value is greater than the no-load closure threshold and less than the lower limit of the effective gripping range, or greater than the upper limit of the effective gripping range and less than the lower limit of the complete release tolerance range, the gripping state is determined to be abnormal. The robotic arm is controlled to reset and the visual pose calculation of step S1 is re-executed until the grasp is successful or the number of retries reaches the preset retry limit, and then an abnormal alarm is triggered.

[0087] In this embodiment, the protocol-level control range of the position pulse of the servo motor of the clamping part is 0 to 100, and the actual effective motion range is measured and calibrated to be 1 to 97. The position pulse feedback value of 1 corresponds to the clamping part being fully closed, and the position pulse feedback value of 97 corresponds to the clamping part being fully open. The no-load fully closed position pulse identifier value is 1, and the no-load closed threshold value is 2 (= no-load fully closed value 1 + preset tolerance 1). The effective clamping range is measured and calibrated to be 70 to 85, with its center value being approximately 77, corresponding to the average position pulse feedback value when the outer diameter of the earpiece is stably clamped. The complete release tolerance range is 96 to 97. The above values ​​are based on the measured reading of the position pulse feedback value of the servo motor of the clamping part of the myCobot 280 robotic arm. When the position pulse feedback value is less than or equal to 2, it is determined that the gripper has closed completely without contacting the earpiece. When the position pulse feedback value is in the range of 70 to 85, it is determined that the earpiece is stably clamped. When the position pulse feedback value is in the range of 3 to 69 or 86 to 95, it is determined that the clamping state is abnormal.

[0088] The position pulse three-state determination logic can be formalized as the following piecewise function:

[0089]

[0090] in, This is the position pulse feedback value of the servo motor of the clamping part; The no-load closing threshold; For the effective clamping interval; To fully release the lower bound of the tolerance range, the above three-state determination mechanism achieves reliable identification of the grasping state under the condition of no torque sensor through pure position servo feedback.

[0091] This step uses the inherent position pulse feedback value of the servo motor as the basis for grasping confirmation, without the need for additional electrical components, and meets the intrinsic safety requirements for mining applications.

[0092] After successful capture confirmation, the edge computing node 102 controls the robotic arm to align the receiving end and the transmitting end of the earpiece 115 and insert them into the first groove structure 118 and the second groove structure 119 of the acoustic interaction base 108. This docking posture is the target posture described by the fixed posture parameters that are optimized in the offline calibration stage and directly reused in the online operation stage.

[0093] The edge computing node 102 acquires the call audio signal collected by the pickup device 109 in the first groove structure 118 of the acoustic interaction base 108, and uses an adaptive filter to suppress background noise on the signal. The reference noise input of the adaptive filter is the background noise signal of the dispatch room environment collected by the independently set environmental microphone 111.

[0094] After suppression processing, edge computing node 102 extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters, calculated using the following formulas:

[0095]

[0096] in, Power estimation of the target speech segment in the call audio signal collected by the pickup device within the first groove structure of the acoustic interaction base. The reference noise power estimate is obtained by adaptive filtering of the background noise signal of the dispatch room environment collected by the environmental microphone.

[0097] The Speech Transmission Index (STI) is calculated based on the Modulation Transfer Function (MTF) principle and is defined as follows:

[0098]

[0099] in, The weighted average of the apparent signal-to-noise ratio for each octave band is calculated using the following formula:

[0100]

[0101] In the formula, The number of octaves is used in this embodiment, which has 7 octaves and center frequencies of 125Hz, 250Hz, 500Hz, 1000Hz, 2000Hz, 4000Hz and 8000Hz respectively. For the first Weighting coefficients for each octave; For the first The average apparent signal-to-noise ratio corresponding to each modulation frequency within each octave band is calculated using the following formula:

[0102]

[0103] In the formula, In this embodiment, four modulation frequencies are used within each octave: 1.0Hz, 2.0Hz, 4.0Hz, and 8.0Hz. For the reference signal at the 1st The first octave Modulation depth at each modulation frequency; To receive the signal at the The first octave The modulation depth at each modulation frequency. The modulation depth is obtained by performing a Hilbert transform on the bandpass-filtered signal to extract the envelope, and then performing a Fast Fourier Transform to obtain the ratio of the amplitude at the corresponding modulation frequency to the mean of the envelope. The value ranges from 0 to 1, with a larger value indicating higher clarity of voice transmission.

[0104] The above The calculation is a simplified engineering implementation based on the modulation transfer function principle, covering 7 octaves, with 4 modulation frequencies per octave, and is used for relative comparison in the closed-loop optimization scenario of this application.

[0105] This embodiment sets a preset signal-to-noise ratio threshold. Preset resolution threshold The above thresholds are calibrated based on the measured values ​​of acoustic quality parameters after optimization and convergence in the offline calibration phase and the measured statistical results of the stability of fixed pose reuse in the online operation phase, to ensure that the fixed pose parameters determined by offline calibration can stably meet the minimum acoustic quality requirements required for speech recognition when directly reused in the online operation phase.

[0106] This step corresponds to the acoustic quality judgment logic in the online operation phase. The pose optimization loop is not executed in the online operation phase. The fixed pose parameters of the end-effector coordinate system of the robotic arm have been predetermined and persistently stored in the offline calibration phase. The complete process of the offline calibration phase is described later.

[0107] After extracting the acoustic quality parameters in step S3, the edge computing node 102 performs a simple compliance judgment:

[0108] If the signal-to-noise ratio is not lower than a preset signal-to-noise ratio threshold Furthermore, the speech clarity index is not lower than a preset clarity threshold. If the acoustic coupling quality is deemed satisfactory, proceed to step S5.

[0109] If the signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold or the speech intelligibility index is lower than the preset intelligibility threshold, the acoustic coupling quality is deemed substandard, the subsequent automatic recognition process is terminated, and the current communication link is transferred to a manual dispatch terminal. An event chain record carrying a marker indicating that manual review is required is generated, and an event marker suggesting recalibration is generated. The event marker suggesting recalibration is used to trigger an offline calibration process to redetermine the fixed pose parameters. After the manual transfer, the edge computing node 102 controls the robotic arm to maintain the clamping state and remove the earpiece 115 from the first groove structure 118 and the second groove structure 119 of the acoustic interaction base 108. The earpiece 115 moves to a preset handover pose according to a preset trajectory and remains stationary, waiting for the manual dispatcher to retrieve it. Handpiece 115 is moved; the preset handover posture is a fixed spatial position that is pre-set so that the human dispatcher can directly pick up the handpiece 115; during the waiting period, the edge computing node 102 controls the clamping part 113 to maintain the preset gripping torque and cyclically reads the position pulse feedback value of the clamping part servo motor 114. If the position pulse feedback value drops below the preset no-load closing threshold (that is, the gripper closes to a near-completely closed position when there is no handpiece blocking it), it is determined that the handpiece 115 has been taken by the human dispatcher. Then, the clamping part 113 is fully opened and the robotic arm is controlled to retreat to the standby posture. In the above-mentioned handover process, the handpiece 115 is not put back into the phone stand 116 to avoid pressing down the phone stand fork spring to trigger the physical hang-up and cause the call link to be accidentally disconnected.

[0110] The aforementioned compliance judgment mechanism eliminates the need for pose optimization loops during online operation. The robotic arm moves directly to the grasping position based on fixed pose parameters and then performs acoustic quality compliance judgment, avoiding response delays caused by successive optimizations. This ensures that the optimal acoustic coupling state guaranteed by the offline calibration phase is achieved immediately upon receiving a call.

[0111] The following is the offline calibration process, which is used to predetermine fixed pose parameters. The offline calibration phase is performed during the idle period when the system is not responding to incoming calls, and the fixed pose parameters determined therein are directly reused in the online operation phase.

[0112] The edge computing node 102 controls the robotic arm to move to the position of the handset 115 for visual pre-alignment. The visual features of the preset reference mark 120 are extracted by the image acquisition device 112. Since the handset 115 is placed on the phone stand 116 at this time, the relative positional relationship between the phone stand 116, the phone body and the reference mark 120 is fixed and known. Therefore, after calculating the pose of the phone body, the pose of the handset 115 relative to the coordinate system of the end tool of the robotic arm is indirectly obtained through the coordinate transformation of the offline calibration. The current pose is recorded as the baseline coordinates. The baseline coordinates are used as the comparison reference in the subsequent coordinate restoration protection judgment.

[0113] Since the mine dispatch telephone has no open electrical interface, it is impossible to establish a call path by triggering the off-hook via electrical signal during the offline calibration phase. The acoustic signal source required for calibration must be provided after the call path is established.

[0114] The edge computing node 102 automatically identifies and prompts a calibrable time window based on historical call frequency statistics. After manual confirmation by the dispatcher, a call is manually initiated from another phone to the phone to be calibrated. After the phone to be calibrated rings, the clamping part 113 is controlled to grasp the handset 115 and establish a call path. Then, the robotic arm is controlled to align the receiving end and the transmitting end of the handset 115 and insert them into the first groove structure 118 and the second groove structure 119 of the acoustic interaction base 108, respectively.

[0115] After the call path is established and the handset 115 is docked with the acoustic interaction base 108, the calling party's phone plays a pre-recorded reference audio. This reference audio serves as the input signal for the acoustic quality parameter extraction in step S3. It is played through the receiving end of the handset 115 of the phone to be calibrated, and collected by the pickup device 109 within the first groove structure 118 of the acoustic interaction base 108 before being input into the acoustic quality parameter extraction process. The aforementioned reference audio setting ensures that the acoustic coupling quality parameters optimized during the calibration phase are consistent in signal characteristics with the acoustic quality parameters determined during the online operation phase. The optimization target during the offline calibration phase is the pose of the robotic arm when docking the handset 115 with the acoustic interaction base 108.

[0116] In the heuristic pose optimization strategy based on coordinate descent with directional continuation, the initial step translation displacement is set to 20mm; the minimum step translation displacement is 10mm. When the value is lower than this, the change in acoustic quality parameters will be lower than the measurement fluctuation threshold and cannot be detected. The single deflection amount of the fixed-length angle deflection is 5 degrees.

[0117] The measured data shows that within the alignment deviation range of 0 to 100 mm, the average decrease slope of the signal-to-noise ratio (SNR) with increasing spacing is approximately -0.11 dB / mm. This means that the SNR change corresponding to every 10 mm spacing is approximately 1.1 dB, which is significantly greater than the fluctuation threshold of the acoustic measurement system (approximately 0.5 dB). Setting the initial step size to 20 mm ensures that the changes in acoustic quality parameters caused by each fine adjustment are clearly discernible, avoiding misjudgment of direction due to the superposition of robotic arm positioning errors and measurement noise. The minimum step size is limited to 10 mm because the SNR change caused by every 10 mm displacement is approximately 1.1 dB, exceeding the detectable range of the fluctuation threshold of 0.5 dB. Below this step size, the change will approach the measurement noise level and cannot be used as a reliable basis for determining the validity of the direction. The step size halving mechanism is triggered when no valid direction can be found in all axes at the current step size. After decreasing from 20 mm to 10 mm, the minimum step size is reached and will not be halved further, avoiding the step size from decreasing to an undetectable scale.

[0118] After the speech path is established and the reference audio is played, the edge computing node activates a coordinate descent heuristic pose optimization strategy based on directional continuation. The target pose of the robotic arm's end-effector coordinate system after each fine-tuning is obtained by the following formula:

[0119]

[0120] Pose fine-tuning vector The rotational magnitudes of the six joint servos of the myCobot280 robotic arm are mapped onto the inverse kinematics solution. This is the homogeneous transformation matrix of the target pose after fine-tuning; This is the homogeneous transformation matrix of the current pose before fine-tuning; Let be the translation transformation matrices for translating along the X, Y, and Z axes of the end-tool coordinate system by dx, dy, and dz, respectively. drx, dy, and dz are rotation transformation matrices representing the angles drx, dry, and drz around the X, Y, and Z axes of the end-tool coordinate system, respectively; dx, dy, and dz are the translational fine-tuning amounts along the X, Y, and Z axes, respectively, in millimeters; drx, dry, and drz are the angular deflection fine-tuning amounts around the X, Y, and Z axes, respectively, in degrees.

[0121] The coordinate descent optimization strategy with directional continuation is executed alternately in three stages: search, continuation, and direction switching. In the search stage, the system performs step translation or fixed-length angle deflection along the preset axis in both directions, and re-extracts acoustic quality parameters after each fine adjustment. When a certain direction is determined to be valid, the system enters the continuation stage and continues to step along that direction until it becomes invalid. When continuation fails, the system enters the direction switching stage, first attempting to revert to the opposite direction, and then switching to the next search axis. If no valid direction can be found in either direction, the step size is halved (after halving, the step size is not less than the preset minimum step translation displacement of 10mm) and the system returns to the search stage for a more refined search. Since acoustic quality parameters include two indicators: signal-to-noise ratio (SNR) and speech intelligibility index (SCI), which may exhibit non-monotonic changes, this embodiment uses the following AND logic to confirm the validity of the fine-tuning direction: when neither the SNR nor the SCI decreases after a single fine-tuning, and the increase in at least one indicator is greater than a preset fluctuation threshold, the fine-tuning direction is confirmed as valid, and the fine-tuning direction that makes both the SNR and SCI exhibit monotonically non-decreasing behavior is retained for further iteration; when the continuation of a certain direction is determined to be invalid, a direction switching phase is entered to try other axes in turn; when no new valid direction can be found in the positive and negative directions of all axes under the current step size, it is determined whether the current acoustic quality parameters have met the requirement that the SNR is not lower than the threshold. And the speech clarity index is no less than If the condition is met, convergence is determined and the current pose is locked. If the condition is not met, the step size is halved (after halving, the step size is not less than the preset minimum step translation displacement of 10mm) and the search phase is returned to continue searching. The above iteration and convergence determination process continues until convergence is determined or the number of iterations reaches the preset upper limit.

[0122] The AND dual threshold direction validity determination logic can be formalized as follows:

[0123]

[0124] in, This represents the change in signal-to-noise ratio before and after fine-tuning. This represents the change in speech clarity index before and after fine-tuning; The signal-to-noise ratio fluctuation threshold is set to 0.5 dB in this embodiment; The threshold value for speech clarity index fluctuation is 0.01 in this embodiment. In the above formula, the first two conjunctive conditions ensure that neither index decreases beyond the fluctuation threshold, and the third disjunctive condition ensures that the increment of at least one index significantly exceeds the fluctuation threshold.

[0125] The convergence criterion is as follows:

[0126]

[0127] in, To preset the signal-to-noise ratio threshold, To preset the speech clarity index threshold, the optimization is considered to have converged only when both acoustic quality indicators simultaneously meet their respective thresholds.

[0128] Considering the engineering implementation The calculation may return invalid values ​​due to signal quality or numerical stability issues. The optimization strategy is equipped with a degradation mechanism: when... When the indicators are unmeasurable or invalid, both the directional validity determination and the convergence determination degenerate into a single indicator mode based solely on the signal-to-noise ratio (SNR). That is, the directional validity criterion is based solely on whether the SNR increment exceeds the fluctuation threshold, and the convergence criterion is based solely on whether the SNR meets the threshold. This degradation mechanism ensures that the system can still complete pose optimization when some acoustic indicators fail, thus possessing engineering robustness.

[0129] If the iteration converges, the current pose is locked as the candidate optimal pose; if the iteration count reaches the preset upper limit and convergence is still not achieved, the pose data with the best acoustic quality parameters in each iteration is used as the candidate optimal pose.

[0130] The acoustic quality parameters corresponding to the candidate optimal pose and the baseline coordinates are compared: if the acoustic quality parameters of the candidate optimal pose are lower than those corresponding to the baseline coordinates, the current calibration optimization result is determined to be inferior to the initial visual pre-alignment result, the system is restored to the baseline coordinates and the calibration is marked as failed, and the fixed pose parameters are not updated; if the acoustic quality parameters of the candidate optimal pose are better than or equivalent to those corresponding to the baseline coordinates, the candidate optimal pose is locked as the fixed pose parameters corresponding to the earpiece.

[0131] The aforementioned coordinate restoration protection mechanism can prevent offline calibration from writing pose parameters that are inferior to the initial visual pre-alignment results due to optimization divergence or abnormal fluctuations, thereby ensuring the engineering integrity of fixed pose parameters.

[0132] After successful calibration, the fixed pose parameters are persistently stored in the non-volatile memory of the edge computing node 102, and a mapping relationship between the handset identifier (left or right phone) and the fixed pose parameters is established, so that the corresponding fixed pose parameters can be selected according to the ringing source phone determined by S11 during the online operation phase.

[0133] The above calibration procedure was performed independently on both sides of the dispatch console, and the fixed pose parameters were recorded separately. The fixed pose parameters of the two sides of the phones were independent of each other to avoid coupling quality deviations caused by differences in the physical positions of the two sides of the phones.

[0134] After generating an event flag suggesting recalibration during the online operation phase, the edge computing node 102 identifies a calibrable time window based on historical call frequency statistics and provides a prompt. After manual confirmation by the dispatcher, the operator manually initiates a call to the phone to be calibrated from another phone to trigger the offline calibration process; or the offline calibration process is executed in response to the dispatcher's active triggering instruction.

[0135] To verify the mapping relationship between acoustic coupling quality parameters and the alignment deviation of the receiver 115 relative to the pickup device 109 of the first groove structure 118 of the acoustic interaction base 108, a gap scanning experiment was conducted under laboratory conditions using an independent test bench with adjustable spacing. In the experiment, a standard reference audio was played at a fixed power. The signal-to-noise ratio and speech intelligibility index were extracted after being acquired by the pickup device 109 fixed on the independent base. Scanning measurements were performed within an alignment deviation range of 0 to 100 mm. The measured data are as follows:

[0136]

[0137] The above measured data reveals two key characteristics of the acoustic coupling quality parameters:

[0138] First, the signal-to-noise ratio (SNR) shows a monotonically decreasing trend as the alignment deviation increases, dropping from 28.45 dB at a deviation of 0 mm to 17.01 dB at a deviation of 100 mm. The average decreasing slope is approximately -0.11 dB / mm, meaning that the SNR change corresponding to every 10 mm spacing is approximately 1.1 dB. This is significantly greater than the fluctuation threshold of the acoustic measurement system (approximately 0.5 dB), constituting a physical constraint on the minimum detectable step size. The aforementioned slope is calculated from large-pitch scanning data rather than being directly measured through millimeter-level step sizes.

[0139] Secondly, the speech intelligibility index exhibits a significant non-monotonic variation as the alignment deviation increases, decreasing from 0.9519 at a deviation of 0 mm to a minimum of 0.6540 at a deviation of 40 mm, then rising again to 0.9244 at a deviation of 70 mm, before decreasing again to 0.7745 at a deviation of 100 mm. This non-monotonic variation stems from reverberation, standing waves, and multipath effects of acoustic signals under near-field coupling, confirming that a single acoustic index cannot stably characterize coupling quality. Therefore, the use of an AND dual-threshold convergence logic combining signal-to-noise ratio and speech intelligibility index in this application is physically necessary.

[0140] Based on the aforementioned acoustic-machine mapping characteristics, a coordinate descent heuristic pose optimization strategy with directional continuation was executed under laboratory conditions using a standard reference audio as the calibration signal. The experiment was conducted on an independent test bench in the laboratory. A pre-recorded standard reference audio file was used to simulate the reference speech played by the caller's phone to verify the effectiveness of the acoustic-machine mapping relationship and the AND dual-threshold convergence logic. This was not based on a real mining dispatch telephone call scenario; the measured background noise was -54.7 dB. The following are the measured data for key steps in the optimization iteration process (to clearly demonstrate the direction switching and convergence decision logic, only the key steps of the initial determination of validity in each direction and direction switching are listed in the table; intermediate steps of continuous steps along the same direction in the continuation phase are omitted):

[0141]

[0142] The convergence was achieved better than step 3. At convergence, SNR = 20.41 dB (above the threshold of 18.0 dB) and STI = 0.8369 (above the threshold of 0.50). From the initial state to the convergence state, SNR improved by 3.12 dB and STI improved by 0.3143. It should be noted that although SNR and STI both reached their respective thresholds at the end of step 1, after the Z-direction continuation failed, the system tried the RZ+ and other axes in sequence during the direction switching phase until no new valid direction could be found in all axes under the current step size. At this time, the convergence judgment was triggered. By this time, step 3 had been iterated, so the convergence was judged to be in step 3. This process reflects the execution sequence of "exploring first when switching directions and judging convergence after exhausting the search space".

[0143] The measured data from the above iterative process confirms the necessity of the AND dual-threshold convergence logic: In step 1, the signal-to-noise ratio (SNR) increases significantly by 1.65 dB while the speech intelligibility index decreases slightly by 0.0021. If only the speech intelligibility index is used as a single indicator, the fine-tuning direction will be misjudged as invalid. In step 3, the SNR increases by 0.97 dB while the speech intelligibility index decreases slightly by 0.0043. This small decrease is within the preset fluctuation threshold range and should still be considered as a valid direction. This non-monotonic variation characteristic makes the single-indicator optimization strategy risky in terms of direction misjudgment. This application effectively avoids this risk by using the AND logic of two indicators.

[0144] The above measured data was obtained based on a laboratory alternative method, which involved using the second speaker to play audio as a reference to simulate a real incoming call from the caller's extension, rather than verifying a real incoming call. Future on-site verification data should be supplemented in a real mining telephone environment when conditions permit.

[0145] To verify the engineering feasibility of directly reusing the fixed pose parameters determined by offline calibration during the online operation phase, the online operation process was repeatedly executed 15 times in a complete mode (including actual grasping / releasing / manual picking up and placing of the earpiece) under laboratory conditions. Each time, the machine directly moved to the grasping position based on the fixed pose parameters, without executing a pose optimization loop. The measured data are as follows:

[0146] Signal-to-noise ratio statistics: the mean was 19.86 dB, the standard deviation was 0.70 dB, the minimum was 18.15 dB, the maximum was 21.28 dB, and the range was 3.13 dB. The signal-to-noise ratio of all 15 measurements was higher than the preset signal-to-noise ratio threshold of 18.0 dB, and the compliance rate was 100%.

[0147] Speech clarity index statistics: mean value 0.7745, standard deviation 0.1351, minimum value 0.4716, maximum value 0.9440, range 0.4724. Among them, the speech clarity index of 14 measurements was higher than the preset clarity threshold of 0.50, with a compliance rate of 93.3%; the measurement value of 0.4716 was lower than the threshold, triggering the manual transfer and suggesting recalibration.

[0148] Statistics on the position pulse feedback values ​​of the servo motor of the clamping part (reflecting the physical reset accuracy and gripping consistency of the earpiece holder): the mean value is 77.1, the standard deviation is 1.2, the minimum value is 74, the maximum value is 79, and the range is 5. This range data indicates that the earpiece holder can still maintain stable physical reset accuracy after repeated picking and putting. The position pulse feedback values ​​are concentrated in the effective clamping range of 70 to 85, which confirms the engineering feasibility of directly reusing fixed posture parameters in the online operation phase.

[0149] Based on the above three statistics, in 15 complete online operation tests, the proportion of both signal-to-noise ratio and speech clarity index meeting the standards simultaneously was 93.3%, meaning that the success rate of directly reusing fixed pose parameters in the online operation phase was 93.3%. This test result verifies the engineering effectiveness of the two-stage architecture that separates offline calibration and online operation: in the online operation phase, there is no need to execute a pose optimization loop to achieve the acoustic coupling quality that meets the requirements of speech recognition in most cases. Only a small number of abnormal situations caused by physical reset deviation of the earpiece seat are handled through manual transfer and suggested recalibration marking mechanism, which reflects the core advantage of no optimization delay in the online operation phase and the security of the degradation mechanism.

[0150] The above measured data were also obtained based on a laboratory alternative, namely, using the second speaker to play standard reference audio to simulate the actual incoming call from the caller's extension, rather than verifying the actual incoming call.

[0151] The edge computing node 102 transmits the call audio signal, whose acoustic quality parameters have reached the quality threshold, to the physically isolated local computing node 104 via the local area network 103 in the mining area. The physically isolated local computing node 104 performs localized speech recognition and voiceprint recognition to obtain the call text content and caller identity information.

[0152] The physically isolated local computing node 104 obtains the incoming call extension number and corresponding fixed work location information, retrieves the currently stored shift schedule data, and performs a three-factor consistency comparison of the incoming call identity information, the incoming call extension number, and the shift schedule data. Automatic response permissions for security commands are granted only when all three match completely; otherwise, the communication link is transferred to a manual dispatch console.

[0153] The physically isolated local computing node 104 matches the call text content and caller identity information with the pre-stored coal mine safety regulations knowledge base to generate response audio data. This response audio data is then sent to the edge computing node 102 via the mine's internal local area network 103. The response audio is played through the miniature speaker 110 in the second groove structure 119 of the acoustic interaction base 108. The response audio is played towards the handset 115 via the miniature speaker 110 fixed to the inner wall of the second groove structure 119. The dispatcher on the dispatch exchange side or the other party in the call can receive the response audio.

[0154] After the response playback is completed, a second-level logic hang-up verification is triggered: within a preset time window after the response playback is completed, the pickup device 109 in the first groove structure 118 of the acoustic interaction base 108 continuously monitors to detect the busy tone characteristics of the communication network and human voice characteristics. After the preset time window expires, the edge computing node 102 controls the robotic arm to maintain the clamping state and remove the earpiece 115 from the first groove structure 118 and the second groove structure 119 of the acoustic interaction base 108.

[0155] The edge computing node 102 controls the robotic arm to move the handset 115 to above the phone stand 116 along a preset trajectory, and controls the clamping part 113 to perform a release action, triggering the first-level physical hang-up verification: the image acquisition device 112 confirms that the handset 115 has entered the preset return area of ​​the phone stand 116, and detects that the position pulse feedback value of the clamping part servo motor 114 has reached the preset release tolerance range, then the physical hang-up action is determined to be completed. If the physical hang-up action is completed and the second-level logical hang-up verification condition has been met within the preset time window, a normal closed-loop event chain record is generated; if the physical hang-up action is completed but the second-level logical hang-up verification condition has not been met, an event chain record carrying an abnormality mark to be reviewed is generated; if the physical hang-up action is not completed, the hang-up action is re-executed (i.e., the handset 115 is moved back to above the phone stand 116 and the clamping part 113 is released), until the physical hang-up action is completed or the number of retries reaches the preset limit.

[0156] In the above three-stage judgment, the degradation processing path of "if the physical hang-up action is completed but the previous logical hang-up verification conditions are not met, an event chain record with an abnormality mark to be reviewed is generated" reflects the degradation fault tolerance principle based on the physical hang-up result: even if the logic layer fails to confirm the termination of the call through busy tone detection or voice timeout, as long as the physical layer confirms that the handset has been returned to its original position, the hang-up action is considered to be completed, avoiding phone deadlock and busy line.

[0157] The above-mentioned cascading verification mechanism is as follows: First, a second-level logical hang-up verification is performed to detect the call status. After the preset time window expires, the acoustic interaction base is withdrawn and a first-level physical hang-up verification is performed to confirm that the earpiece is physically back in place. If the logical hang-up verification fails but the physical hang-up verification passes, an abnormal record to be reviewed is generated based on the physical hang-up result. This mechanism realizes hang-up confirmation without electrical feedback at both logical and physical levels, avoiding the deadlock and busy line problem caused by relying solely on disconnection tone verification under conditions without electrical feedback.

[0158] The multimodal cascaded hang-up verification mechanism can be formalized as the following cascaded decision logic:

[0159]

[0160] in, The result of the first-level physical hang-up verification is obtained by logically ANDing the two sub-conditions: visual repositioning confirmation and position pulse release confirmation.

[0161]

[0162] In the formula, This is a visual positioning indicator function. It takes the value of true when the handset enters the preset positioning area of ​​the handset dock, and false otherwise. This is a position pulse release indication function. It is true when the position pulse feedback value of the servo motor of the clamping part reaches the preset release tolerance range, and false otherwise.

[0163] The result of the second-level logical hang-up check is obtained by logically ORing the two sub-conditions: busy tone feature detection and human voice timeout detection.

[0164]

[0165] In the formula, This is a busy tone feature indication function. It takes the value of true when the audio signal collected by the pickup device detects the busy tone feature of the communication network, and is false otherwise. The human voice timeout indicator function is set to true when the duration of consecutive absence of human voice features exceeds a preset threshold, and false otherwise. This cascaded decision structure embodies the principle of "logic-level first, physical-level confirmation" for degraded fault-tolerant design.

[0166] This invention is essentially a system-level solution addressing the stringent technical constraints of mining areas lacking electrical interfaces and requiring network isolation. Its advantages are as follows:

[0167] First, by combining purely visual and acoustic degradation-cascaded hang-up confirmation, the deadlock problem under conditions without electrical feedback is overcome. Under conditions without electrical feedback signals, relying solely on the execution of mechanical actions without introducing a multimodal state confirmation mechanism will result in misjudgment of the hang-up status and telephone deadlock / busy line problems. This invention uses visual positioning confirmation and position pulse feedback to form the first level of physical hang-up verification, and uses busy tone feature extraction and human voice timeout detection to form the second level of logical hang-up verification. It also sets up a degradation processing path based on the physical level results, thus realizing hang-up confirmation under conditions without electrical feedback at both physical and logical levels.

[0168] Second, at the edge, a gripper servo position pulse feedback offset torque sensor is used to achieve non-tactile gripping confirmation. The two-stage architecture of offline calibration and online operation is used to overcome the problem of blind gripping and sound leakage caused by insufficient physical alignment accuracy of passive acoustic devices. In the offline calibration stage, acoustic coupling quality parameters (signal-to-noise ratio and speech intelligibility index) are used as teaching targets, and the reference audio played after the voice path is established is used as the teaching signal. AND double threshold convergence logic is used to solve the problem of direction validity determination under non-monotonic acoustic mapping, and a coordinate restoration protection mechanism is set to ensure the engineering integrity of fixed pose parameters. In the online operation stage, fixed pose parameters are directly reused to avoid response delay caused by successive optimization. The experimental data of this invention show that the Speech Intelligibility Index (STI) exhibits a non-monotonic change (first decreasing and then increasing) during the alignment deviation scanning process of the earpiece receiving end relative to the first groove structure pickup device of the acoustic interaction base. This indicates that there are local non-monotonic characteristics in the acoustic coupling quality space, so a single index gradient cannot be used as the sole optimization basis. The AND dual threshold convergence logic of signal-to-noise ratio (SNR) and speech intelligibility index (STI) is a specific adaptation to this physical characteristic, enabling stable determination of directional validity even under non-monotonic acoustic mapping.

[0169] Third, an asymmetric computing power allocation mode of real-time physical control at the edge and high-computing-power semantic processing at the local level was constructed in terms of architecture. Edge nodes are responsible for low-computing-power real-time control, while local computing power nodes are responsible for high-computing-power business logic. The two communicate through a physically isolated local area network, which satisfies the data security isolation requirements of the mining area while completing the reasonable division of computing power tasks.

[0170] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for answering mine telephone calls without electrical interfaces based on a pure position servo robotic arm, employing a distributed communication architecture including edge computing nodes of the robotic arm and physically isolated local computing nodes in the mining area. An acoustic interaction base is fixedly installed within the working range of the robotic arm. The acoustic interaction base has a first groove structure and a second groove structure. A sound pickup device is installed on the inner wall of the first groove structure, and a miniature speaker is installed on the inner wall of the second groove structure. The method is characterized in that... The method includes the following steps performed in response to an incoming call event: Ring detection steps: The edge computing node acquires an image of the phone's ring indicator light through an image acquisition device and acquires an ambient audio signal through an ambient microphone. It performs ringing spectrum feature recognition on the ambient audio signal and ring indicator light status recognition on the image. When both determine that a ringing event exists, it identifies the phone from which the ringing occurs and triggers the off-hook process. Off-hook and gripping steps: The edge computing node selects the corresponding pre-determined fixed posture parameters according to the phone from which the ringing is coming from, controls the robotic arm to move to the handset position according to the fixed posture parameters, controls the end gripping part to perform a closing action, and determines the gripping state based on the position pulse feedback value of the gripping part servo motor; Acoustic docking steps: After successful grabbing, control the robotic arm to align the receiver and transmitter of the earpiece and insert them into the first groove structure and the second groove structure respectively; Acoustic quality judgment steps: The edge computing node acquires the call audio signal collected by the pickup device, extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters, and judges whether the acoustic quality parameters are lower than the preset quality threshold. Voice recognition and response steps: If the acoustic quality parameters are not lower than the quality threshold, the call audio signal is transmitted to the physically isolated local computing node via the mine's internal LAN for localized voice recognition and voiceprint recognition to obtain the call text content and caller identity information. The call text content and caller identity information are matched with the pre-stored coal mine safety regulations knowledge base to generate response audio data, which is then sent to the edge computing node via the mine's internal LAN and played through a miniature speaker. Hang-up verification steps: After the response playback is completed, the second-level logical hang-up verification is triggered to detect the call termination status. The robotic arm is controlled to remove the earpiece from the first groove structure and the second groove structure. Then, the first-level physical hang-up verification is triggered to confirm that the earpiece is physically back in place. An event chain record is generated based on the results of the second-level logical hang-up verification and the first-level physical hang-up verification.

2. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, The gripping state is determined based on the position pulse feedback value of the servo motor of the gripping part, specifically: If the pulse feedback value at this position is less than or equal to the preset no-load closing threshold, the grabbing is determined to have failed. If the pulse feedback value at this position is within the preset effective gripping range, the grasp is considered successful. If the pulse feedback value at this position is greater than the no-load closing threshold and less than the lower limit of the effective clamping interval, or greater than the upper limit of the effective clamping interval and less than the lower limit of the complete release tolerance interval, then the clamping state is determined to be abnormal. When the grasping fails or the clamping state is abnormal, the control robot arm is reset and the unloading and grasping steps are re-executed until the grasping is successful or the number of retries reaches the preset retry limit, then an abnormal alarm is triggered. The preset no-load closure threshold is the position pulse feedback value of the servo motor that deviates from the fully closed position of the clamping part within a preset tolerance range. The preset effective clamping interval is a continuous interval within a preset tolerance range centered on the average position pulse feedback value of the servo motor when the outer diameter of the earpiece is stably clamped.

3. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 2, characterized in that, The quality thresholds include a preset signal-to-noise ratio threshold and a preset clarity threshold. In the acoustic quality judgment step, if the signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold or the speech clarity index is lower than the preset clarity threshold, the subsequent automatic recognition process is stopped, the current communication link is transferred to the manual dispatch terminal, an event chain record carrying a manual review mark is generated, and an event mark suggesting recalibration is generated. After being transferred to the manual dispatch terminal, the edge computing node controls the robotic arm to maintain the clamping state, remove the earpiece from the first groove structure and the second groove structure, and move to the preset handover posture to wait for the manual removal of the earpiece. During the waiting period, the position pulse feedback value of the servo motor of the clamping part is read cyclically. If the position pulse feedback value drops below the no-load closing threshold, it is determined that the earpiece has been removed, and the clamping part is controlled to open and retreat to the standby position.

4. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, The fixed pose parameters are predetermined through the following offline calibration steps: Control the robotic arm to move to the earpiece position for visual pre-alignment, and record the current pose as baseline coordinates; The call is initiated manually from another phone to the phone to be calibrated. After the phone rings, the clamping part grabs the handset and the handset is picked up to establish a call path. The robotic arm aligns the receiving end and the transmitting end of the handset with the first groove structure and the second groove structure respectively. The calling phone plays the pre-recorded reference audio. Activate the coordinate descent heuristic pose optimization strategy based on directional continuation, and sequentially control the robotic arm to perform step translation along the preset axis of the end-effector coordinate system or deflect at a fixed angle around the preset axis. After each fine adjustment, the acoustic quality parameters are re-extracted. When the signal-to-noise ratio and speech intelligibility index do not decrease after a single fine adjustment, and the increase of at least one indicator is greater than the preset fluctuation threshold, the fine adjustment direction is confirmed to be effective and retained, and the iteration continues until the signal-to-noise ratio is not lower than the preset signal-to-noise ratio threshold and the speech intelligibility index is not lower than the preset intelligibility threshold, or the number of iterations reaches the preset upper limit. If the acoustic quality parameters are still lower than the preset quality threshold when the preset upper limit is reached, the pose data with the best acoustic quality parameters in each iteration is called as the candidate optimal pose. The acoustic quality parameters corresponding to the candidate optimal pose and the baseline coordinates are compared. If the acoustic quality parameters of the candidate optimal pose are lower than the acoustic quality parameters corresponding to the baseline coordinates, the baseline coordinates are restored and the calibration is marked as failed. Otherwise, the candidate optimal pose is locked to the fixed pose parameters corresponding to the earpiece and stored persistently. The above calibration process was performed independently on the two telephones on both sides of the dispatch console, and the fixed posture parameters were recorded respectively.

5. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, After the local computing node performs localized speech recognition and voiceprint recognition to obtain the call text content and caller identity information, it also includes a three-factor joint identity verification: the local computing node obtains the caller extension number and the corresponding fixed work location information, obtains the current shift schedule data stored locally, and performs a consistency comparison of the caller identity information, the caller extension number and the shift schedule data. Only when the three are completely matched will the automatic response permission involving security commands be granted; otherwise, the communication link will be transferred to the manual dispatch console.

6. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, Before the edge computing node extracts the signal-to-noise ratio and speech intelligibility index of the call audio signal as acoustic quality parameters, it also includes: using an adaptive filter to suppress background noise in the call audio signal, wherein the reference noise input of the adaptive filter is the background noise signal of the dispatch room environment collected by an environmental microphone independently set outside the mine dispatch telephone.

7. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, The local area network within the mining area is a dedicated network that connects to the mining area's intranet access point via wireless communication modules deployed on edge computing nodes, or via a direct network cable connection through a switch. Furthermore, there is no data routing or exchange between the physically isolated local computing nodes and the external public internet.

8. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, Before controlling the robotic arm to move to the earpiece position according to the fixed pose parameters, the edge computing node also compares the calculated current pose of the earpiece with the fixed pose parameters to confirm that the earpiece is still in the expected position.

9. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 1, characterized in that, In the hang-up verification step, the second-level logical hang-up verification is as follows: within a preset time window after the response playback is completed, the microphone continuously monitors to detect the busy tone characteristics of the communication network and the characteristics of human voice. The first-level physical hang-up verification is as follows: the image acquisition device confirms that the handset has entered the preset return area of ​​the handset base, and the position pulse feedback value of the clamping part servo motor is detected to reach the preset complete release tolerance range.

10. The method for answering a mining telephone without an electrical interface based on a pure position servo robotic arm as described in claim 9, characterized in that, An event chain record is generated based on the results of the second-level logical hang-up verification and the first-level physical hang-up verification, specifically: If the first-level physical hang-up verification passes and the second-level logical hang-up verification conditions are met, then a normal closed-loop event chain record is generated. If the first-level physical check passes but the second-level logical check fails, an event chain record with an exception flag to be reviewed is generated. If the first-level physical AFK verification fails, the AFK action will be re-executed until the first-level physical AFK verification passes or the number of retries reaches the preset limit.

Images