An electric power-assisted delivery box based on individualized multi-physiological signal fusion and short-distance private crowdsourcing and an adaptive control method thereof

Abstract

The application discloses an electric power-assisted delivery box based on individualized multi-physiological signal fusion and short-distance private crowdsourcing and an adaptive control method thereof. The bottom of the box is provided with a detachable electric power-assisted module through a guide rail, which comprises a hub motor, an electromagnetic clutch and a polyurethane friction wheel; a handle controller integrates a PPG heart rate sensor and a thin film pressure grip sensor, and is connected to a box control unit through a waterproof bus; the control unit fuses heart rate, heart rate variability and grip to calculate an individualized fatigue index based on an individualized physiological baseline, and calculates adaptive power output in combination with slope and load; meanwhile, the control unit receives platform road condition early warning through a 4G communication module, or directly exchanges inertial measurement unit data with surrounding boxes through a Bluetooth low-power Mesh short-distance communication module to realize low-delay road condition crowdsourcing, and adopts an anonymization identifier and a local automatic deletion mechanism to protect the privacy of operators. The application realizes a delivery aid of instant use, individualized anti-overwork, real-time road condition sharing and offline autonomy.

Description

Technical Field

[0001] This invention belongs to the technical field of electric-assisted cargo transport equipment, specifically relating to an electric-assisted cargo box installed on the rear rack of a bicycle or electric bicycle, an adaptive assist control method based on the fusion of individualized physiological baselines of the operator and multi-source heterogeneous signals, and a short-range privacy-preserving road condition crowdsourcing system between cargo boxes based on Bluetooth Low Energy Mesh. This invention can be widely applied to scenarios such as food delivery, express delivery, postal delivery, and mobile vending. Background Technology

[0002] In the on-demand delivery industry, operators commonly use non-powered insulated boxes or ordinary cargo containers for transporting goods. With increasingly stringent delivery time requirements, operators often work continuously for more than 8 hours a day, highlighting the growing conflict between accumulated physical fatigue and driving safety. Existing technologies in this field have the following technical shortcomings: (1) Existing vehicle power steering solutions do not take into account the cumulative fatigue of the operator and have technical biases. Current electric bicycle power assist systems generally rely on speed, cadence, or gradient signals, employing universal control thresholds. This approach implicitly assumes that "all users have the same physiological endurance under identical working conditions," failing to consider the cumulative fatigue effect of operators delivering goods continuously for more than 8 hours. When the operator is fatigued, the system continues to provide full assistance, causing the operator to ride at high speeds with diminished reaction time, posing a serious safety hazard. Furthermore, these power assist systems are expensive and fixed to the bicycle, making them unremovable and failing to meet the needs of delivery riders for flexible bike switching and quick assembly / disassembly. (2) Existing wearable devices do not form a closed loop with vehicle assistance, and the use of a common threshold leads to a high false alarm rate. While existing photoplethysmography (PPG) heart rate monitoring technology has been applied in smart bracelets and car seats, it is only used for health reminders or displays and does not form a closed-loop control system with vehicle power assist systems. More importantly, existing solutions generally use a universal heart rate threshold, failing to consider individual differences: the resting heart rate of operators of different ages and physical conditions can vary by more than 20 bpm. This "one-size-fits-all" approach leads to a high rate of false alarms or missed alarms due to fatigue. In addition, existing wearable devices are not optimized for cycling scenarios and do not consider motion artifacts on bumpy roads, resulting in a significant decrease in the reliability of heart rate data on uneven surfaces. (3) Existing road condition perception relies on the cloud, which has large latency and poor privacy. Current mobile navigation systems rely on cloud-based maps to update road conditions, with update delays typically ranging from 5 to 15 minutes. This makes them unable to detect centimeter-level bumps and vibrations, such as those caused by construction steel plates or speed bumps. Some solutions propose cellular network communication between containers, but cellular networks require relaying through base stations, preventing direct broadcasting between devices. Communication delays between containers are typically greater than 500ms, failing to meet the requirements for low-latency road condition sharing. Furthermore, existing solutions upload GPS coordinates and inertial measurement unit data to the cloud platform, posing a risk of operator location privacy leaks. The existing technology suffers from a technical bias that "road condition information must be centrally processed in the cloud before it can be shared," hindering the realization of low-latency direct collaboration between containers. (4) Existing technologies lack modular, low-cost solutions for delivery boxes. Modifying wheel assist requires specialized tools and typically takes over 30 minutes to install; furthermore, the vehicle's power assist system is separate from the cargo box, making it impossible to provide targeted assistance to the cargo box. For scenarios such as food delivery that require frequent loading and unloading of cargo boxes and vehicle changes, the existing solutions are too costly in terms of both installation and time. In summary, the existing technologies suffer from the following technological biases that urgently need to be overcome: First, the technical bias that "a universal physiological threshold applies to all users" leads to fatigue detection failure in delivery scenarios with large individual differences. Second, the technical bias that "road condition information must be uploaded to the cloud for centralized processing" makes it impossible for road condition information to be shared between the boxes with low latency and poses a high risk of privacy leakage. Third, the inherent perception that "PPG signals must be used under static or low-motion-interference conditions" leads to the direct discarding of heart rate signals in high-bump cycling scenarios instead of intelligent repair. This invention is proposed in response to the aforementioned technical biases and defects. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide an electric-assisted delivery box based on individualized multi-physiological signal fusion and short-distance privacy crowdsourcing and its adaptive control method, so as to solve the problems of existing delivery boxes having no assistance, assistance not taking into account individual fatigue differences and baseline drift, lagging road condition perception, unreasonable communication architecture between boxes and insufficient privacy protection. The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: An electric-assisted delivery box, installed on the rear rack of a bicycle or electric bicycle, includes a box body, a detachable electric assist module, a load detection module, a physiological signal acquisition module, a control unit, and an individualized physiological baseline calibration module. The electric assist module is located at the bottom of the box body and is detachably connected to the box body via guide rails. It includes a hub motor, an electromagnetic clutch, and a friction wheel. The physiological signal acquisition module includes a PPG heart rate sensor and a thin-film pressure grip sensor located within the handlebar controller. The handlebar controller is connected to the control unit via a waterproof bus. The control unit includes a main control chip, a six-axis inertial measurement unit, a barometer, a 4G communication module, and a short-range communication module. The individualized physiological baseline calibration module stores the operator's individualized physiological baseline and automatically corrects baseline drift according to a preset period. This invention also provides an adaptive control method, comprising seven steps: baseline calibration, signal acquisition and artifact elimination, fatigue determination, adaptive assist calculation, road condition crowdsourcing, assist correction, and safety protection. The present invention also provides a road condition crowdsourcing electric-assisted system based on short-range privacy communication, comprising multiple electric-assisted delivery boxes as described above, wherein the boxes communicate directly with each other via a Bluetooth Low Energy Mesh network to achieve anonymous and localized road condition sharing. In the above technical solution, there is an inseparable synergistic relationship among the four core technical means: First, individualized physiological baseline calibration provides an individual reference standard for fatigue assessment through three-signal fusion. Without an individualized baseline, relative heart rate load, relative heart rate variability, and relative grip strength cannot be calculated, and three-signal fusion will degenerate into a general threshold assessment, leading to a significant increase in the false alarm rate. Second, the IMU-assisted PPG motion artifact elimination algorithm ensures the continuity of three-signal fusion in bumpy cycling scenarios. Without this algorithm, the PPG signal fails in high-interference regions, and the three-signal fusion degenerates into a single signal (grip force) determination, resulting in a significant decrease in sensitivity. Third, the three-signal fusion fatigue assessment provides accurate physiological state input for adaptive assist output. Compared to the single-signal scheme, the three-signal fusion achieves multi-dimensional cross-validation based on an individualized baseline, reducing the fatigue misjudgment rate to less than one-third of that of the single-signal scheme. Fourth, BLE Mesh short-range privacy crowdsourcing provides forward-looking road condition input for adaptive assistance. This method, in conjunction with the previous three methods, enables the assistance output to not only respond to the current physiological state but also to anticipate road conditions ahead, forming a dual closed-loop control of "physiological adaptation + road condition prediction". The above four technical means support each other and are indispensable, together constituting the complete technical solution of the present invention. Implementing any one of the means alone cannot solve the technical problem of "individualized, low-latency, and highly reliable adaptive assistance and safety protection in delivery scenarios" that the present invention aims to solve. Compared with existing technologies, the present invention achieves the following unexpected technical effects: (1) The accuracy of individualized fatigue determination is significantly improved: Through three-signal fusion and individualized baseline calibration, the accuracy of fatigue determination can still be maintained above 85% in a long-term test of continuous use for 30 days; while the general threshold scheme without individualized baseline and drift correction has an accuracy of about 70% during the same period. (2) Significantly reduced heart rate estimation error in high interference scenarios: By switching between IMU triaxial auxiliary filtering and grip force-dominant switching, the heart rate estimation error can be controlled within 5% in the high interference range where vertical acceleration is greater than 1.5 times the gravitational acceleration, which is significantly better than the scheme that directly uses the original PPG data (the error is usually greater than 15%). (3) The delay of road condition sharing between the boxes is greatly reduced: Through the direct connection between BLE Mesh boxes, the communication delay between the boxes is controlled within 50ms, which is an order of magnitude lower than the delay of cellular network equipment forwarded through the base station (usually greater than 500ms), meeting the requirement of early warning 3 seconds in advance. (4) Operator location privacy is substantially protected: Through anonymization and local automatic deletion mechanism, the amount of operator location data uploaded is reduced by more than 90%, and local area network data is only retained locally for a preset time before being deleted and not uploaded to the cloud. (5) Significantly improved network stability in high-density scenarios: Through the adaptive broadcast interval mechanism, the broadcast conflict rate can be reduced to below 5% in scenarios with a node density of more than 15 nodes per 300 meters; while without the adaptive mechanism, the conflict rate can reach more than 30%. (6) Improved energy recovery efficiency for extended driving range: Under typical delivery conditions (70% on flat roads, 20% uphill, 10% downhill, average assist power 60-80W, load 15kg, continuous for 8 hours), the remaining power can reach more than 15%; under the condition of a 15° downhill slope of 500m, the recovered power can reach more than 10% of the total power consumption. Attached Figure Description Figure 1 This is a schematic diagram of the overall appearance of the electric-assisted delivery box of the present invention; Figure 2 This is an exploded view of the bottom of the housing of the present invention; Figure 3 This is a cross-sectional view of the motor clutch mechanism of the present invention, showing the energized engagement state and the de-energized disengagement state; Figure 4 This is a schematic diagram of the handle controller of the present invention; Figure 5 This is a hardware connection topology diagram of the system of the present invention; Figure 6 This is a flowchart illustrating the overall process of the adaptive assist control method of the present invention. Figure 7 This is a flowchart illustrating the individualized fatigue index calculation logic of the present invention. Figure 8 This is a schematic diagram of the dual-mode communication architecture of the present invention; Figure 9 This is a sequence diagram of the local area network traffic crowdsourcing of the present invention; Figure 10 This is a state transition diagram of the security protection mechanism of the present invention; Figure 11 This is a flowchart of the PPG motion artifact removal algorithm of the present invention; Figure 12 This is a schematic diagram of the node strategy for the BLE Mesh adaptive broadcast mechanism of the present invention; Figure 13 This is a bar chart comparing the effects of the present invention with those of existing technologies; Figure 14 This is a flowchart of the baseline drift correction algorithm of the present invention; Figure 15 This is a schematic diagram of the anonymization identifier generation mechanism of the present invention.

Claims

1. An electric-assisted delivery box, installed on the rear rack of a bicycle or electric bicycle, characterized in that, include: Main body of the box; An electric power assist module is located at the bottom of the main body of the casing and is detachably connected to the main body of the casing via a guide rail. The electric power assist module includes a hub motor, an electromagnetic clutch, and a friction wheel. The electromagnetic clutch is configured to: engage when energized, driving the friction wheel to overcome the elastic force of the return spring and press against the rear tire of the bicycle; and when de-energized, the return spring drives the friction wheel to separate from the tire. The load detection module is located at the bottom of the main body of the box and includes multiple weighing sensors for detecting the load on the box. The physiological signal acquisition module includes a PPG heart rate sensor and a membrane pressure grip sensor installed in the handle controller; the handle controller is connected to a control unit installed in the main body of the housing via a waterproof bus; The control unit includes a main control chip, a six-axis inertial measurement unit (IMU), a barometer, a 4G communication module, and a short-range communication module; the control unit is electrically connected to the physiological signal acquisition module, the electric power assist module, and the load detection module, and is configured as follows: (a) Based on the individualized physiological signals acquired by the physiological signal acquisition module and the load detected by the load detection module, a drive signal for controlling the output torque of the electric power assist module is generated; (b) When the vertical acceleration detected by the six-axis inertial measurement unit exceeds the preset interference threshold, the PPG data of the corresponding time period is marked as a high interference interval, and linear extrapolation estimation of the preceding normal heart rate data is adopted, or the grip strength-dominated mode is switched to temporarily increase the weight of grip strength in fatigue determination. (c) Upload inertial measurement unit data and receive road condition warnings through the 4G communication module, or directly exchange inertial measurement unit data with surrounding boxes on the same platform within a preset distance through the short-range communication module; The individualized physiological baseline calibration module is located inside the main body of the box and is configured to store the operator's individualized physiological baseline and automatically correct baseline drift at a preset cycle during continuous use; The individualized physiological baselines include resting heart rate, maximum heart rate, grip strength baseline, and heart rate variability baseline.

2. An adaptive control method for electric power assistance based on individualized multi-physiological signal fusion and short-distance privacy crowdsourcing, applied to the electric power assistance delivery box described in claim 1, characterized in that, Includes the following steps: S1: Baseline calibration step: The control unit collects and stores the operator's physiological baseline through the individualized physiological baseline calibration module. The physiological baseline includes resting heart rate, maximum heart rate, grip strength baseline and heart rate variability baseline. The maximum heart rate is estimated based on the operator's age and supports calibration based on actual exercise peak measurements; S2: Signal Acquisition and Artifact Removal Steps: During cycling, the control unit samples the heart rate signal through the PPG heart rate sensor and the grip force signal through the membrane pressure grip force sensor. After filtering, the real-time heart rate, heart rate variability and grip force are determined. When the vertical acceleration detected by the six-axis inertial measurement unit exceeds the preset interference threshold, the corresponding time period is marked as a high interference interval, and linear extrapolation estimation is performed using previous normal heart rate data, or the grip strength-dominated mode is switched. S3: Fatigue determination step: The control unit calculates an individualized fatigue index based on the real-time heart rate, heart rate variability, grip strength and the physiological baseline. The individualized fatigue index is based on a weighted fusion of relative heart rate load, relative heart rate variability change and relative grip strength change. When the individualized fatigue index exceeds the preset fatigue threshold, it is determined to be a fatigue state, and a first adjustment signal is generated; S4: Adaptive Assist Calculation Step: The control unit responds to the operator's manual gear position, load, slope, and the first adjustment signal to calculate the target assist output value and generate a target control signal to drive the hub motor; the slope is obtained by calculating the height change rate of the barometer. S5: Road Condition Crowdsourcing Step: The control unit uploads inertial measurement unit data through the 4G communication module and receives road condition warning information generated based on regional crowdsourcing data; or directly exchanges inertial measurement unit data with surrounding boxes through the short-range communication module to calculate the local turbulence prediction index. S6: Assist correction step: The control unit generates a second adjustment signal based on the road condition warning information or the local bump prediction index to adjust the target control signal; S7: Safety protection steps: When the heart rate is detected to exceed the preset proportion of the maximum heart rate or be lower than the preset minimum heart rate, or the fatigue index exceeds the preset fatigue threshold for more than the first preset duration, or the grip strength is lower than the preset proportion of the grip strength baseline for more than the second preset duration, the control unit executes the corresponding safety protection action.

3. A road condition crowdsourcing electric assist system based on short-range privacy communication, characterized in that, Includes multiple electrically assisted delivery boxes as described in claim 1, with each delivery box communicating directly with the others via a Bluetooth Low Energy Mesh network; The control units for each delivery box are configured as follows: Anonymous identifiers, location coordinates, and standard deviation of vertical acceleration of the inertial measurement unit within a preset time period are broadcast via Bluetooth Low Energy Mesh. Receive broadcast data from other boxes within a preset distance and calculate the local turbulence prediction index by weighting the distance exponential attenuation. When the local turbulence prediction index exceeds the preset threshold and there is a turbulence source within a preset distance ahead, the motor output torque is pre-adjusted and a buzzer is sounded to alert the system. Local area network data is automatically deleted after a preset retention period and is not uploaded to the cloud.

4. The electric-assisted delivery box according to claim 1, characterized in that, The handle controller has a built-in independent safety microcontroller unit, which is connected to the control unit via the waterproof bus and is configured to send an emergency braking command directly to the electric power assist module when the grip force is detected to be lower than a preset ratio for more than a second preset time. The control unit is also configured to perform the following safety protection actions: When the heart rate exceeds 0.85 times the maximum heart rate or falls below 40 bpm, disconnect the power assist and report to the platform; When the fatigue index exceeds 0.7 for more than 60 seconds, the assist limit will be reduced and a buzzer alarm will sound. If the grip strength is below 0.3 times the grip strength baseline for more than 10 seconds, emergency braking is triggered and the incident is reported to the platform. When the battery level is below 20%, the assist limit is reduced.

5. The electric-assisted delivery box according to claim 1, characterized in that, It also includes a generator-assisted braking module, wherein the control unit is configured to: when the downhill slope S < -0.05 is detected by the barometer, control the hub motor to switch to generator mode, provide auxiliary braking torque and recover the generated current to the battery; The braking strategy of the generator-assisted braking module is graded as follows: Only energy is recovered on a slight downhill slope (-0.05>S≥-0.10); Energy recovery combined with light braking is used on moderate downhill slopes (-0.10>S≥-0.20); Energy recovery is superimposed on forced motion on steep slopes (S<-0.20).

6. The method according to claim 2, characterized in that, In step S2, the determination of the high interference interval satisfies any of the following conditions: The vertical acceleration a_z is greater than 1.5 times the gravitational acceleration; The absolute value of the rate of change of the resultant acceleration is greater than 3 times the gravitational acceleration per second; The periodicity of the triaxial acceleration overlaps more than 60% with the spectrum of the PPG signal; The formula for calculating the individualized fatigue index F in step S3 is as follows: F = w_hr × HR_rel + w_hrv × HRV_rel + w_grip × Grip_rel in: HR_rel = (HR - HR_rest) / (HR_max - HR_rest), which represents the relative heart rate load; HRV_rel = max(0, 1 - HRV / HRV_base), which represents the relative heart rate variability. Grip_rel = max(0, 1 - Grip / Grip_base), which represents the relative grip strength variation; HR is the real-time heart rate, HR_rest is the resting heart rate, HR_max is the maximum heart rate, HRV is the real-time heart rate variability, HRV_base is the baseline of heart rate variability, Grip is the real-time grip strength, and Grip_base is the grip strength baseline. w_hr, w_hrv, and w_grip are the heart rate weight, heart rate variability weight, and grip strength weight, respectively, and w_hr + w_hrv + w_grip = 1; In normal mode, w_hr = 0.5, w_hrv = 0.3, w_grip = 0.2; In grip strength-dominated mode, w_grip = 0.5, w_hr = 0.3, w_hrv = 0.

2.

7. The method according to claim 2, characterized in that, The method for calculating the adaptive assist output in step S4 is as follows: Calculate the base assist value T_base = k_m × M × W, where k_m is the gear coefficient, M is the operator's manual gear, and W is the load. The slope compensation assist value T_slope = T_base × (1 + 2×S) is calculated, where S is the slope calculated by the barometer height change rate. The fatigue compensation boost value T_fatigue = T_slope × (1 + 0.5×F) is calculated, where F is the individualized fatigue index; Calculate the boost value after bump reduction: T_bump = T_fatigue × (1 - 0.3×B_warning), where B_warning is the bump warning indicator, which is 1 when a road condition warning is received or the local bump prediction index exceeds 0.25g, and 0 otherwise. Determine the final boost output T_final = min(T_bump, 250W); When the vehicle speed exceeds 25km / h, T_final returns to zero; when the individualized fatigue index F>0.7, the upper limit of T_final drops to 150W; when the battery charge is below 20%, the upper limit of T_final drops to 100W.

8. The method according to claim 2, characterized in that, The specific steps in step S5, including exchanging data with surrounding enclosures via a short-range communication module, include: The device sends its anonymized identifier, location coordinates, and standard deviation of vertical acceleration of the inertial measurement unit within the past 10 seconds via Bluetooth Low Energy Mesh broadcast; the anonymized identifier is generated by hashing the device identifier, current timestamp, and random salt value, and is changed every 10 minutes. Receives broadcast data from other units on the same platform within a 300m radius; The local turbulence prediction index is calculated using a distance exponential decay weighting method, with the weighting function being e^(-d / 100), where d is the distance to the surrounding enclosures. Local area network data is automatically deleted after being cached locally for 30 seconds and is not uploaded to the cloud. The broadcast adopts an adaptive broadcast interval strategy: when the number of surrounding nodes is less than 5, the broadcast interval is 100ms; when the number of surrounding nodes is between 5 and 15, the broadcast interval is 200ms; when the number of surrounding nodes is greater than 15, it only unicasts to the 3 nearest nodes. The method also includes an offline autonomous mode: when no 4G network signal is detected, only the short-range communication module is activated to exchange data with the surrounding enclosures, or only the auxiliary output is calculated based on physiological signals; when the 4G network is restored, the inertial measurement unit data stored during the offline period is synchronized in batches. The batch synchronization adopts differential uploading, and only abnormal data with σ_a>0.2g are uploaded.

9. The method according to claim 2, characterized in that, The baseline drift correction algorithm in step S1 is as follows: Let the current baseline be B_old, and the average value of the newly collected data over the past 7 days be B_new; Modified formula: B_updated = α × B_new + (1-α) × B_old; Where α is the correction factor, and its value ranges from 0.1 to 0.3; Adjust the range limit: |B_updated - B_old| / B_old ≤ 10%; If the operator's resting heart rate changes by more than 15% from the baseline for three consecutive days, automatic correction will be paused and a recalibration prompt will be displayed.

10. The system according to claim 3, characterized in that, The Bluetooth Low Energy Mesh network adopts an adaptive broadcast interval: when the number of surrounding nodes is less than 5, the broadcast interval is 80ms-120ms; when the number of surrounding nodes is between 5 and 15, the broadcast interval is 180ms-220ms; when the number of surrounding nodes is greater than 15, it only unicasts to the nearest 3 nodes. The anonymization identifier is generated as follows: the input device physical address, current timestamp, and 32-byte random salt value are used, and the first 8 bytes are taken as the anonymization identifier after SHA-256 hash operation; if a hash collision occurs, a random number is appended and the hash is recalculated.

Images