Robot obstacle avoidance method and device based on adaptive shared control

Abstract

This invention relates to the field of robot control technology, and more particularly to a robot obstacle avoidance method and apparatus based on adaptive shared control. The method includes: constructing a robot state vector and constructing an operator behavior input sequence; calculating orientation scores and behavior scores, and structurally combining the orientation scores and behavior scores to generate a confidence index; introducing a robot autonomous control output sequence, and simultaneously mapping the operator behavior sequence to control signals in the execution control command space, fusing the output sequence and control signals, and outputting a fused control signal; introducing an adsorption-type speed adjustment model and a safety suppression factor, adjusting the fused control signal through the adsorption-type speed adjustment model, and generating an actual execution control signal by combining the safety suppression factor and the adjusted signal, and sending the actual execution control signal to the motor controller through the robot control panel, and the motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and more particularly to a robot obstacle avoidance method and apparatus based on adaptive shared control. Background Technology

[0002] With the increasing prevalence of robots in industrial logistics, public services, medical assistance, and special operations, the obstacle avoidance capability of mobile robots in complex and dynamic environments has become a key factor influencing system performance. Traditional mobile robots largely rely on fully autonomous environmental perception and path planning mechanisms to perform obstacle avoidance tasks. Their control strategies typically construct an environmental model using LiDAR, encoders, and inertial measurement units, and then generate control commands based on local or global planners. However, in highly uncertain scenarios such as confined spaces, crowded areas, and areas with dense dynamic obstacles, autonomous control is often limited by sensor blind spots, delayed environmental information updates, and conservative path selection. This can lead to sluggish obstacle avoidance responses, overly conservative operations, or trajectories that fail to meet task objectives, and even passive stalling due to planning failures when approaching obstacles. To overcome the limitations of autonomous obstacle avoidance, human-robot shared control is increasingly seen as a collaborative mode that enhances system flexibility and fault tolerance. In this mode, operators can intervene in real-time with the robot through control devices, thereby compensating for sensor limitations or proactively avoiding potential risks.

[0003] However, existing shared control schemes often simply integrate human input with autonomous planning instructions through fixed ratios or rule-based logic, lacking a systematic assessment of operator behavior stability, intent rationality, and environmental risk levels. Therefore, it is difficult to form a reliable dynamic control switching mechanism. When operators make temporary errors, experience sudden changes in directional instructions, or experience high-frequency input fluctuations, such fixed strategies can easily lead to discontinuous or even abrupt changes in the robot's trajectory, increasing the risk during obstacle avoidance and reducing the overall smoothness and safety of the system. Furthermore, some studies have attempted to incorporate learning models to infer operator intent, but the model output is usually difficult to directly combine with the robot's underlying control structure for stable execution, making it difficult to achieve controllable, interpretable, and continuous collaborative obstacle avoidance actions in real systems. Therefore, there are still significant technical gaps in constructing human-robot shared obstacle avoidance systems with high safety, high responsiveness, and natural control transitions, especially requiring a dynamic control mechanism that can simultaneously handle short-term operator behavior trends, input rationality, and changes in environmental risk. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a robot obstacle avoidance method and apparatus based on adaptive shared control.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Robot obstacle avoidance methods based on adaptive shared control include:

[0007] A robot state vector is constructed based on the robot's sensor data, and an operator behavior input sequence is constructed based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the obstacle in front.

[0008] The operator's behavior sequence is mapped to the intention angle, the difference between the current orientation angle and the intention angle is calculated, the difference and the minimum distance to the forward obstacle are input into the direction scoring function to generate a direction score, and the behavior score is generated by the behavior scoring function based on the operator's behavior sequence. The direction score and behavior score are structurally combined to generate a confidence index.

[0009] The robot autonomous control output sequence is introduced, and the operator behavior sequence is mapped to control signals in the execution control command space. The robot autonomous control output sequence and control signals are fused by a fusion function to output a fused control signal. The parameters of the fusion function include confidence index, dynamic damping factor and conflict enhancement factor.

[0010] An adsorption-type speed adjustment model and a safety inhibition factor are introduced. The fused control signal is adjusted by the adsorption-type speed adjustment model. The actual execution control signal is generated by combining the safety inhibition factor and the adjusted signal. The actual execution control signal is sent to the motor controller through the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles.

[0011] Preferably, the generation of the current orientation angle includes:

[0012] By reading the wheel assembly rotation information from the encoder and combining it with the linear acceleration and angular velocity provided by the IMU, the current orientation angle of the robot's current position in the two-dimensional plane is obtained through conventional kinematic derivation.

[0013] Preferably, the operator behavior sequence includes the operator's control tendencies in the horizontal and vertical directions.

[0014] Preferably, the parameters of the direction scoring function include an angle smoothing factor, an enhancement factor, and the minimum distance to the forward obstacle.

[0015] Preferably, the parameters of the behavior scoring function include a penalty coefficient and an operator behavior sequence, wherein the penalty coefficient is used to control the sensitivity to continuous changes.

[0016] Preferably, the parameters of the conflict enhancement factor include a normalization factor, a magnitude, and a conflict gain coefficient. The normalization factor is used to represent the degree of directional opposition between human-machine control signals, the magnitude is used to control the input difference, and the conflict gain coefficient is used to adjust the conflict sensitivity.

[0017] Preferably, the safety suppression factor includes a configurable minimum safety distance threshold, a minimum obstacle distance, and a suppression factor.

[0018] Preferably, the adsorption-type velocity adjustment model includes a fusion control signal, the current velocity, and a scaling factor.

[0019] Preferably, after adjusting the fusion control signal using the adsorption-based velocity adjustment model, the method further includes:

[0020] Calculate the deviation between the adjusted signal and the robot's current operating state;

[0021] If the deviation exceeds the threshold, the adjusted signal is corrected, and the degree of pullback of the proportional coefficient is controlled.

[0022] A robot obstacle avoidance device based on adaptive shared control includes:

[0023] The state synchronization modeling module is used to construct a robot state vector based on the robot's sensor data and to construct an operator behavior input sequence based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the forward obstacle.

[0024] The confidence assessment module is used to map the operator's behavior sequence to an intention angle, calculate the difference between the current orientation angle and the intention angle, input the difference and the minimum distance to the forward obstacle into the direction scoring function to generate a direction score, and generate a behavior score based on the operator's behavior sequence through the behavior scoring function. The direction score and behavior score are structurally combined to generate a confidence index.

[0025] The signal fusion module is used to introduce the robot's autonomous control output sequence and map the operator's behavior sequence into control signals in the execution control command space. The robot's autonomous control output sequence and control signals are fused together by a fusion function to output a fused control signal. The parameters of the fusion function include confidence index, dynamic damping factor and conflict enhancement factor.

[0026] The obstacle avoidance action output module is used to introduce an adsorption-type speed adjustment model and a safety suppression factor. The fused control signal is adjusted through the adsorption-type speed adjustment model, and the actual execution control signal is generated by combining the safety suppression factor and the adjusted signal. The actual execution control signal is sent to the motor controller through the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles.

[0027] The beneficial effects of this invention are as follows:

[0028] This invention introduces structured modeling of robot motion states and operator input data, and incorporates analysis of the operator's short-term control behavior trends and directional intentions. Figure 1 A comprehensive analysis mechanism considering consistency and current environmental risks generates a real-time adjustment factor reflecting operator credibility, serving as the core parameter for the shared control phase. Guided by this parameter, this invention further constructs a damped control fusion strategy for shared control scenarios. This strategy coordinates human input and autonomously planned control outputs nonlinearly, ensuring smooth and controllable control signals even in typical shared control situations such as directional conflict, forced deflection, or rapid takeover. Furthermore, this invention designs a speed buffer and safety suppression mechanism at the execution end, tailored to the actual chassis dynamics. This allows the fused control commands to be executed continuously while satisfying dynamic constraints, and improves obstacle avoidance safety through nonlinear deceleration when obstacles approach. The innovation of this invention lies in introducing adaptive control adjustment logic based on behavioral credibility, constructing a human-machine control fusion structure that combines stability and flexibility, and forming a unified closed loop with the dynamic safety constraints at the execution end. This solves problems such as delayed shared control response, abrupt control signal changes, and discontinuous obstacle avoidance behavior in existing technologies, enabling robots to achieve safer, more natural, and more operator-intended obstacle avoidance actions in complex dynamic environments. Attached Figure Description

[0029] Figure 1 This is a flowchart of a robot obstacle avoidance method based on adaptive shared control in a specific embodiment of the present invention;

[0030] Figure 2 This is a block diagram of a robot obstacle avoidance device based on adaptive shared control in a specific embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] refer to Figure 1 As shown, this application proposes a robot obstacle avoidance method based on adaptive shared control, including:

[0033] Step 1: Construct a robot state vector based on the robot's sensor data, and construct an operator behavior input sequence based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the obstacle in front, specifically including:

[0034] The goal of this step is to build a state input system to support a shared control decision mechanism, specifically one that simultaneously reflects the robot's current dynamic state and the instantaneous input and short-term behavioral trends of human intervention. Unlike traditional path planning systems that only use the robot's environmental state, this step requires incorporating human control behavior into the system state in a structured manner, providing fundamental data support for subsequent human-robot shared control weight decisions. The key to this step lies in how to simultaneously collect these two types of data and construct a behavior cache structure that can be used to determine trends in real time, thus serving as input for confidence assessment in subsequent steps.

[0035] The construction of the robot's state vector relies on standard sensor data from the robot itself, primarily including current position, velocity, and distance information to forward obstacles. Current position and attitude data are acquired by a wheel encoder and an inertial measurement unit (IMU). The encoder reads the wheel rotation information at a frequency of 20 Hz, and combined with the linear acceleration and angular velocity provided by the IMU, the robot's current position coordinates in the two-dimensional plane are derived through conventional kinematics. with orientation angle Speed ​​data, including linear velocity, is returned by the chassis drive module at fixed intervals. With angular velocity Obstacle distance data is provided by LiDAR, which scans a 180-degree range with a 0.5-degree angular resolution. The shortest distance point within the main forward direction (e.g., directly in front ±15 degrees) is selected as the representative value, denoted as . The above three sets of data are integrated to form the robot's state vector. The format is as follows:

[0036] ;

[0037] in The position of the robot in a two-dimensional coordinate system. For the orientation angle, The current linear velocity, The current angular velocity, This represents the minimum obstacle distance in the forward main obstacle avoidance direction. These values ​​are obtained in real time through the robot controller's drive interface and do not involve semantic extraction or complex model calculations.

[0038] Operator historical behavior input vector Provided by a remote control or joystick, it typically includes two channels: one for speed commands in the forward / backward direction and one for steering commands in the left / right direction. The device reads the current input state every 20ms, obtaining a range of... The analog quantity, of which Indicates maximum forward movement without steering. This represents the maximum right turn in place. To model the operator's behavioral trends, this step constructs a fixed-length sequence of operator behaviors. Used to save the past Human input behavior in frames. It is defined as follows:

[0039] ;

[0040] in Indicates a point in time The control input vector at time t, It is a sequence of sliding buffer operator actions maintained by the main control thread. Each time a new frame of data is received, the array is updated, discarding the oldest item and adding the latest frame. Buffer length This can be set according to the system refresh rate, for example, when the refresh rate is 20Hz. This indicates the control behavior over the past 250ms.

[0041] This caching structure can capture typical operational behaviors such as continuous fine-tuning or sudden direction changes, representing a significant enhancement over the traditional method of only collecting single-frame input. Taking a typical scenario as an example, an operator might rapidly and continuously issue left-turn commands to avoid obstacles that the robot has not yet recognized. In a system without historical behavior caching, this command might be judged as a misoperation or an anomaly and directly filtered out. However, with this caching structure… After caching, the system will identify the trend of consecutive operations in the same direction, providing strong support for subsequent shared control weight allocation. This step outputs two state variables: and This data will be fully utilized for the next step of control confidence assessment. All data comes from the robot's standard control interface or the communication module of the control device, ensuring the consistency and real-time performance of the control chain.

[0042] Step Two: Map the operator's behavior sequence to an intention angle, calculate the difference between the current orientation angle and the intention angle, and input the difference and the minimum distance to the forward obstacle into a direction scoring function to generate a direction score. Simultaneously, based on the operator's behavior sequence, generate a behavior score through a behavior scoring function. Combine the direction score and behavior score to generate a confidence index, specifically including:

[0043] The goal of this step is to establish a confidence assessment mechanism for human-robot shared control scenarios in robot obstacle avoidance tasks. This mechanism is used to determine whether the operator's current control input is trustworthy and appropriate to be assigned a high control weight in the current environmental state. Unlike traditional autonomous obstacle avoidance or purely manual control, the system proposed in this application exists in a dynamic intermediate state of "cooperative control." The operator's instructions may be based on prior awareness of dangers that the robot has not yet recognized, or they may be due to misoperation or unstable control behavior. Therefore, a mechanism is needed that considers the timing, intention, and other factors. Figure 1 A confidence generation method that has the ability to discriminate in dimensions such as consistency and environmental dependence.

[0044] The input includes the robot state vector constructed in step one. and operator behavior sequence . It contains the robot's current position information, speed information, and orientation angle. and minimum distance to forward obstacles ; It is by The frame operator's behavior input is an array, with each frame being a two-dimensional vector. These represent the operator's control tendencies in the horizontal (left and right) and vertical (front and back) directions, respectively.

[0045] Unlike traditional strategies, this step proposes a spatial intention-based approach. Figure 1 The confidence generation mechanism, driven by consistency, behavioral stability, and obstacle avoidance urgency, means that confidence is no longer static or empirically set, but dynamically adjusted in real time according to changes in the human-machine relationship.

[0046] This mechanism firstly involves spatial meaning. Figure 1 To improve consistency, a direction difference model is introduced. This involves considering the operator's input at the current moment. Map it to intention angle Then, facing the robot... Calculate the difference Construct a directional scoring function, specifically as follows:

[0047] ;

[0048] in It's a directional score. It is the angle smoothing factor (controls error tolerance). It is an enhancing factor. This represents the distance to the obstacle ahead. The innovation of this formula lies in introducing the obstacle distance into the denominator of the angle error model: when the obstacle is very close, the system's tolerance for directional deviation decreases, thus requiring the operator's direction to be highly consistent with the forward direction in order to obtain high confidence; while when the obstacle is far away or there is no obstacle, the system allows the operator to make a certain degree of deflection to complete autonomous avoidance.

[0049] The second core dimension is the short-term stability of operator behavior. This dimension is used to identify whether the operator is in a state of hesitation, misoperation, or conflicting instructions. The behavior scoring function is defined as follows:

[0050] ;

[0051] in It is a behavioral score. This is the penalty coefficient, representing the control system's sensitivity to continuous rates of change. This formula employs an exponential decay function to ensure that the more drastic the behavioral fluctuations, the greater the impact on confidence. Notably, squared Euclidean distance is used as the input difference term to strengthen the penalty for "reverse tossing" or "high-frequency fine-tuning" behaviors.

[0052] Final confidence index Instead of directly linearly fusing the two scoring terms in the current step, a structure based on confidence-weakening regularization is introduced for combination. This regularization term can be regarded as a threshold control for the system in the low-confidence interval.

[0053] ;

[0054] The advantage of this structure is that when directional consistency is low (i.e. When the value is close to 0, a weakening factor is introduced to prevent the system from retaining high confidence due to a high stability term, thus avoiding the situation of "stable but in the wrong direction". This is a parameter that adjusts the degree of weakening. This structure combines the contributions of the two dimensions through product constraints and nonlinear regularization, which preserves independent contributions while enhancing the ability to adjust for mutual exclusion.

[0055] Through the three-tiered structural design described above, this step implements a mechanism specifically for evaluating the quality of human control input in shared control obstacle avoidance tasks. It does not rely on external training data or complex models, but is entirely based on real-time data acquisition. Variables The value will be used as a direct adjustment factor for the fusion of shared control signals in the next step.

[0056] Step 3: Introduce the robot's autonomous control output sequence, and simultaneously map the operator's behavior sequence into control signals in the execution control command space. A fusion function is used to fuse the robot's autonomous control output sequence and control signals, outputting a fused control signal. The parameters of the fusion function include a confidence index, a dynamic damping factor, and a conflict enhancement factor, specifically including:

[0057] This step aims to construct a fusion control mechanism for human-machine shared control obstacle avoidance tasks. This mechanism uses the confidence index output in step two. As the core regulatory factor, it is input by the operator. With robot autonomous control output Dynamic fusion is performed between them. Unlike traditional static weight control methods, this step designs a damped adjustment model based on the credibility of operator behavior and the system's perceived risk in actual obstacle avoidance scenarios. This ensures that the control signal can respond flexibly and execute stably when dealing with sudden environmental changes, making it particularly suitable for high-dynamic conditions such as high-speed obstacle avoidance and frequent human-machine takeover switching.

[0058] To achieve a realistic and continuous human-machine collaborative control mechanism, two challenges must be addressed: first, how to... The first step involves real-time adjustment of control authority; the second is how to smoothly handle the transition of control signals when there is a significant difference between the operator's and robot's inputs. This step proposes an innovative nonlinear damped control fusion structure based on confidence adjustment, which not only considers the human-machine control input itself but also introduces a penalty term for the degree of input conflict, thereby achieving a more flexible and scenario-adaptive control signal generation method. First, the operator's behavior sequence is... Mapped to control signals in the execution control instruction space This mapping is accomplished through a control interface mapping function, transforming the analog values ​​input from the handle into the linear and angular velocities required by the platform. To maintain the consistency of the control signals, and Maintaining the same data structure. Traditional shared control uses a weighted average method to fuse human-machine control signals. This step introduces a nonlinear conflict penalty term and a damping adjustment term to form the following fusion function:

[0059] ;

[0060] in This is the final output control signal; It is a confidence level indicator; For dynamic damping factor; This is the conflict amplification factor. The third term in this formula is a key innovation, used to dynamically compensate for control deviations caused by differences in human-machine input based on weighted fusion, thus avoiding [further issues]. When the input direction is large but directly controlled by humans, system oscillations occur. Specifically, The design takes the following factors into consideration:

[0061] ;

[0062] In the formula The normalization factor represents the degree of directional opposition between human-machine control signals; that is, when the two signals are in opposite directions, this value tends to amplify the conflict. This controls the magnitude of the input difference, used to measure the degree of absolute difference. The entire conflict enhancement factor... The structural design ensures that when the system faces severe resistance in the control direction, it automatically adds a difference penalty term in the control fusion to reduce the actual weight of human input, thereby ensuring path continuity and system stability. This is the conflict gain coefficient, used to adjust the sensitivity of the mechanism.

[0063] Damping factor The calculation is strongly correlated with the confidence level, and is set as follows:

[0064] ;

[0065] in This is the system's maximum damping constant. This structure ensures that as the confidence level decreases, the damping term increases, thus offsetting the impact of untrusted commands on the fused control results; while... As the value approaches 1, the damping term automatically approaches 0, allowing human control to guide the robot's execution path. Through the above structural design, this step achieves adjustment and enhancement of control signal fusion at multiple levels. Particularly in the confidence boundary region, the system no longer uses fixed linear fusion but introduces a structural dynamic compensation term, making the fusion result more aligned with the human-robot collaborative obstacle avoidance requirements. This mechanism is particularly suitable for the application scenarios defined in this application: complex task environments such as dynamic obstacle avoidance, real-time collaborative control, and unpredictable operator input.

[0066] Step 4: Introduce an adsorption-based speed adjustment model and a safety suppression factor. Adjust the fused control signal using the adsorption-based speed adjustment model. Combine the safety suppression factor and the adjusted signal to generate the actual execution control signal. Send this actual execution control signal to the motor controller via the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles. Specifically, this includes:

[0067] This step focuses on fusing the control signals output from the previous stage. This translates into specific motion commands executable at the robot chassis level, ensuring that the robot can perform safe, stable, and naturally responsive obstacle avoidance maneuvers within a shared control context. Unlike traditional robot control, the fusion signal proposed in this application not only includes a trade-off between the human operator's intention and the autonomous obstacle avoidance outcome, but also introduces a damping term and a conflict penalty mechanism. Therefore, the signal must be further processed during the execution phase to prevent abrupt changes or conflicts with physical constraints, especially to avoid behaviors such as inertial slippage and steering overshoot in highly dynamic environments or areas with multiple obstacles.

[0068] The input for this step is This variable originates from step three, from... (Confidence index) (Human control instructions) (Robot autonomous control instructions) are generated jointly by damping and conflict penalty terms. Typically in two-dimensional vector form , representing linear velocity and angular velocity respectively, with their numerical ranges limited by the robot platform's dynamic capabilities and maximum speed limits. For example, in a wired control platform, Can be set to (unit speed ratio) for (Unit speed ratio). This input structure is consistent with the actual chassis control interface and can be transmitted via serial port or CAN bus, with a typical frequency of 50Hz.

[0069] To ensure dynamic smoothness during motion, this step proposes introducing a velocity inertia constraint mechanism during the execution phase based on the fused signal to prevent... Compared with the current actual speed Excessive discrepancies can cause sudden changes in mechanical stress or attitude instability. In actual deployment, Feedback is received from the chassis controller, typically reported every 20ms via the PID control module inside the drive board. This feedback signal structure is similar to... Same, including and Two components.

[0070] To achieve this mechanism, the system designs an adsorption-based velocity adjustment model, constructed as follows:

[0071] ;

[0072] The core idea of ​​this structure is to fuse control signals. If there is a significant difference between the current speed and the robot's current operating state, the execution signal will be "pulled back" to a speed range closer to the current state, with a proportional coefficient. To control the callback strength, the value is usually set to... Between. For example, if And the current speed The system will gradually transition to the fusion speed instead of jumping abruptly. This formula is simple in structure but highly practical, effectively suppressing the problems of "abrupt turning" or "sudden acceleration" that occur during human-machine control switching, and is especially suitable for narrow passages or frequent obstacle avoidance scenarios.

[0073] In addition to speed buffering, this step also introduces a safety suppression term based on obstacle distance. This factor aims to ensure the robot automatically reduces its speed in critical obstacle avoidance situations, preventing collisions caused by operator error or system fusion errors. The LiDAR module operates at a frequency of 10Hz~20Hz in the system and can provide real-time... The system extracts the main direction (forward direction) from the distance measurement data within the range. Minimum obstacle distance within the range This distance, after software filtering, is used to construct the following suppression term:

[0074] ;

[0075] in It is a configurable minimum safe distance threshold, typically set to [value]. rice, As an inhibitory factor, in Adjustable within a certain range. When an obstacle enters the safe critical distance, this factor will decrease rapidly, multiplying by the control signal to achieve overall deceleration. This mechanism is more flexible than traditional emergency stop strategies and is suitable for shared control scenarios where the operator may be "correcting" the machine's behavior but has not completely avoided the obstacle, allowing the system to gradually intervene and take over.

[0076] Ultimately, the actual control signals executed are given by the following structure:

[0077] ;

[0078] This signal is sent to the motor controller via a standard serial port or CAN protocol through the robot control board (such as STM32, Jetson Orin, or a dedicated chassis driver). The controller then decomposes the signal into the corresponding motor speed or PWM signal to drive the robot to actually turn or move. The execution layer no longer performs any path or control logic judgment; it is only responsible for issuing and executing commands in a closed loop according to the control cycle (e.g., 20ms). After system deployment, in dynamic obstacle avoidance scenarios, such as when an operator wants the robot to quickly move along a wall, multiple rapid turning commands may be generated. This mechanism will then... While granting some control, the system limits its speed to prevent accidental collisions caused by continuous input. Conversely, if the robot's autonomous path is more conservative in obstacle-heavy areas, the system will allow greater control transfer when the operator has high confidence, thus achieving more precise obstacle avoidance closer to human intent.

[0079] refer to Figure 2 As shown, in a second aspect of this application, a robot obstacle avoidance device based on adaptive shared control is proposed, comprising:

[0080] The state synchronization modeling module is used to construct a robot state vector based on the robot's sensor data and to construct an operator behavior input sequence based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the forward obstacle.

[0081] The confidence assessment module is used to map the operator's behavior sequence to an intention angle, calculate the difference between the current orientation angle and the intention angle, input the difference and the minimum distance to the forward obstacle into the direction scoring function to generate a direction score, and generate a behavior score based on the operator's behavior sequence through the behavior scoring function. The direction score and behavior score are structurally combined to generate a confidence index.

[0082] The signal fusion module is used to introduce the robot's autonomous control output sequence and map the operator's behavior sequence into control signals in the execution control command space. The robot's autonomous control output sequence and control signals are fused together by a fusion function to output a fused control signal. The parameters of the fusion function include confidence index, dynamic damping factor and conflict enhancement factor.

[0083] The obstacle avoidance action output module is used to introduce an adsorption-type speed adjustment model and a safety suppression factor. The fused control signal is adjusted through the adsorption-type speed adjustment model, and the actual execution control signal is generated by combining the safety suppression factor and the adjusted signal. The actual execution control signal is sent to the motor controller through the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles.

[0084] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A robot obstacle avoidance method based on adaptive shared control, characterized in that, include: A robot state vector is constructed based on the robot's sensor data, and an operator behavior input sequence is constructed based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the obstacle in front. The operator's behavior sequence is mapped to the intention angle, the difference between the current orientation angle and the intention angle is calculated, the difference and the minimum distance to the forward obstacle are input into the direction scoring function to generate a direction score, and the behavior score is generated by the behavior scoring function based on the operator's behavior sequence. The direction score and behavior score are structurally combined to generate a confidence index. The expression for the directional scoring function is: ； in Score for direction. Angle smoothing factor, As an enhancing factor, Distance to the obstacle ahead. This is the difference between the orientation angle and the intention angle; The behavior scoring function is expressed as follows: ； in It is a behavioral score. It is the penalty coefficient. , The input vector represents the operator's behavior at adjacent time steps, and n is the buffer length. The expression for the confidence index is: ； in As a confidence level indicator, To adjust the parameters for the degree of weakening, Score for direction. Score the behavior; The robot autonomous control output sequence is introduced, and the operator behavior sequence is mapped to control signals in the execution control command space. The robot autonomous control output sequence and control signals are fused by a fusion function to output a fused control signal. The parameters of the fusion function include confidence index, dynamic damping factor and conflict enhancement factor. The fusion function is expressed as follows: ； in For fusion control signals; As a confidence level indicator; For dynamic damping factor; As a conflict amplifying factor, For control signals, This is the output sequence for the robot's autonomous control. An adsorption-type speed adjustment model and a safety inhibition factor are introduced. The fused control signal is adjusted by the adsorption-type speed adjustment model. The actual execution control signal is generated by combining the safety inhibition factor and the adjusted signal. The actual execution control signal is sent to the motor controller through the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles. The adsorption-based velocity adjustment model is expressed as follows: ； in To actually execute control signals, This is the proportionality coefficient. To integrate control signals, This represents the current actual speed.

2. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The generation of the current orientation angle includes: By reading the wheel assembly rotation information from the encoder and combining it with the linear acceleration and angular velocity provided by the IMU, the current orientation angle of the robot's current position in the two-dimensional plane is obtained through conventional kinematic derivation.

3. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The operator behavior sequence includes the operator's control tendencies in the horizontal and vertical directions.

4. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The parameters of the direction scoring function include the angle smoothing factor, the enhancement factor, and the minimum distance to the forward obstacle.

5. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The parameters of the behavior scoring function include a penalty coefficient and an operator behavior sequence, wherein the penalty coefficient is used to control the sensitivity to continuous changes.

6. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The parameters of the conflict enhancement factor include a normalization factor, a magnitude, and a conflict gain coefficient. The normalization factor is used to represent the degree of directional opposition between human-machine control signals, the magnitude is used to control the input difference, and the conflict gain coefficient is used to adjust the conflict sensitivity.

7. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The safety inhibition factors include a configurable minimum safety distance threshold, a minimum obstacle distance, and an inhibition factor.

8. The robot obstacle avoidance method based on adaptive shared control according to claim 1, characterized in that, The adsorption-based speed adjustment model includes a fusion control signal, the current speed, and a scaling factor.

9. The robot obstacle avoidance method based on adaptive shared control according to claim 8, characterized in that, After adjusting the fusion control signal using the adsorption-based velocity adjustment model, the method further includes: Calculate the deviation between the adjusted signal and the robot's current operating state; If the deviation exceeds the threshold, the adjusted signal is corrected, and the degree of pullback of the proportional coefficient is controlled.

10. A robot obstacle avoidance device based on adaptive shared control, characterized in that, include: The state synchronization modeling module is used to construct a robot state vector based on the robot's sensor data and to construct an operator behavior input sequence based on the operator's historical behavior input vector. The robot state vector includes the current orientation angle and the minimum distance to the forward obstacle. The confidence assessment module is used to map the operator's behavior sequence to an intention angle, calculate the difference between the current orientation angle and the intention angle, input the difference and the minimum distance to the forward obstacle into the direction scoring function to generate a direction score, and generate a behavior score based on the operator's behavior sequence through the behavior scoring function. The direction score and behavior score are structurally combined to generate a confidence index. The expression for the directional scoring function is: ； in Score for direction. Angle smoothing factor, As an enhancing factor, Distance to the obstacle ahead. This is the difference between the orientation angle and the intention angle; The behavior scoring function is expressed as follows: ； in It is a behavioral score. It is the penalty coefficient. , The input vector represents the operator's behavior at adjacent time steps, and n is the buffer length. The expression for the confidence index is: ； in As a confidence level indicator, To adjust the parameters for the degree of weakening, Score for direction. Score the behavior; The signal fusion module is used to introduce the robot's autonomous control output sequence and map the operator's behavior sequence into control signals in the execution control command space. The robot's autonomous control output sequence and control signals are fused together by a fusion function to output a fused control signal. The parameters of the fusion function include confidence index, dynamic damping factor and conflict enhancement factor. The fusion function is expressed as follows: ； in For fusion control signals; As a confidence level indicator; For dynamic damping factor; As a conflict amplifying factor, For control signals, This is the output sequence for the robot's autonomous control. The obstacle avoidance action output module is used to introduce an adsorption-type speed adjustment model and a safety inhibition factor. The fused control signal is adjusted through the adsorption-type speed adjustment model. The actual execution control signal is generated by combining the safety inhibition factor and the adjusted signal. The actual execution control signal is sent to the motor controller through the robot control panel. The motor controller decomposes the actual execution control signal to drive the robot to avoid obstacles. The adsorption-based velocity adjustment model is expressed as follows: ； in To actually execute control signals, This is the proportionality coefficient. To integrate control signals, This represents the current actual speed.

Images