A brain-computer interface system and execution method for continuous whole-body control of humanoid robots

Abstract

This invention provides a brain-computer interface system and execution method for continuous full-body control of humanoid robots, belonging to the field of brain-computer interface and robot control technology. The system includes a module for EEG data stream acquisition and online sliding window segmentation, a module for EEG feature extraction and classification, a dual-threshold smoothing module, and a command space mapping and execution module. Specifically, continuous decoding of streaming EEG signals is achieved through online sliding window segmentation with a fixed step size; the confidence rejection and dwell time confirmation mechanism in the dual-threshold smoothing module transforms the unstable original predictions of the decoding model into safe execution commands, effectively filtering invalid EEG signals caused by gaze shifts, blinking, etc., and eliminating command jitter; safe commands are mapped into incremental motion primitives, enabling continuous and precise full-body control of a high-degree-of-freedom humanoid robot. This invention balances low control latency and high stability, significantly improving the safety and control precision of brain-controlled robot interaction.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interface and robot control technology, and relates to a brain-computer interface system and execution method for continuous whole-body control of humanoid robots. Background Technology

[0002] Brain-computer interfaces (BCIs) aim to establish communication and control pathways between humans and external devices that do not rely on peripheral nerves and muscles. In EEG-based motor function assistance and replacement applications, steady-state visual evoked potentials (SSVEPs) have attracted widespread attention due to their advantages such as high information transmission rate and short training cycle.

[0003] Currently, in SSVEP-based mechanical equipment control systems, most research and applications primarily employ a discrete, trial-based control paradigm. In this paradigm, visual stimuli are typically mapped to a finite set of predefined motion commands or fixed target coordinates. The user gazes at the corresponding flashing target area for a fixed duration; the system then extracts the EEG signal during that time period, decodes it using a decoding algorithm, and triggers execution as a discrete action command.

[0004] Existing technical solutions suffer from the following problems: First, discrete control is inherently discontinuous and has high latency. Users must wait for a trial to complete before triggering the next action, failing to meet the robot's need for smooth and rapid response. Second, when transitioning to high-frequency updated continuous control, the shorter time window amplifies the effects of transient noise and inconsistent time patterns, leading to unstable model prediction outputs. Third, in actual closed-loop operation, users inevitably experience brief periods of distraction, blinking, and shifting their gaze away from the stimulus, resulting in blurred or invalid EEG fragments. Executing the decoding results from these periods without discrimination can cause frequent jumps in robot commands or the execution of conflicting actions, introducing safety hazards. Fourth, existing control instruction sets are often too small, typically relying on pre-programmed fixed trajectories, and cannot cover the diverse movements required for humanoid robots with multiple degrees of freedom to perform complex tasks.

[0005] Therefore, there is an urgent need for a brain-computer interface system for continuous full-body control of humanoid robots that can effectively suppress transient noise and invalid signals during continuous decoding, and transform discrete and unstable model predictions into safe and smooth continuous execution instructions, thereby achieving refined and real-time closed-loop control of high-degree-of-freedom robots. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a brain-computer interface system and execution method for continuous whole-body control of humanoid robots, so as to solve the above-mentioned problems existing in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A brain-computer interface system for continuous full-body control of a humanoid robot, the system comprising: EEG data stream acquisition and online sliding window segmentation module: used to continuously segment streaming EEG signals at fixed time steps to generate a series of EEG data windows for real-time decoding; EEG feature extraction and classification module: used to receive the EEG data window, extract steady-state visual evoked potential (SSVEP) features, and output classification scores corresponding to each control command category at each decoding time step; wherein, the classification score is the score vector output by the decoding model for each control command category, and the control command category corresponding to the maximum score in the classification score is defined as the original predicted category, and the original predicted category is used as the candidate command for the current decoding time step; Dual-threshold smoothing module: used to convert the candidate instructions into safe execution instructions. The dual-threshold smoothing module includes: a confidence rejection unit, used to calculate the predicted confidence of the current decoding time step based on the classification score, and when the predicted confidence is lower than a preset confidence threshold, reject the current candidate instruction and output a "no action" instruction; a dwell time confirmation unit, used to determine whether the current candidate instruction has appeared continuously for a preset number of confirmation steps when it is different from the current stable instruction. If the preset number of confirmation steps is reached, the current candidate instruction is confirmed as a new stable instruction and output as a safe execution instruction; otherwise, a "no action" instruction is output. The instruction space mapping and execution module is used to map the safety execution instructions output by the dual-threshold smoothing module into incremental motion primitives of the robot to drive the robot to perform continuous actions; when a "no action" instruction is received, the robot is controlled to maintain the current state.

[0008] Furthermore, in the EEG data stream acquisition and online sliding window segmentation module, the fixed time step is 0.25 seconds, and the EEG data window slides forward at the fixed time step to achieve a command inference frequency of 4Hz.

[0009] Furthermore, in the EEG feature extraction and classification module, the prediction confidence is the maximum probability value of the classification score after normalization by the Softmax function; the confidence rejection unit compares the maximum probability value with the confidence threshold.

[0010] Furthermore, the preset number of confirmation steps in the dwell time confirmation unit is set to 3 steps.

[0011] Furthermore, the instruction space mapping and execution module defines a unified instruction space containing 20 types of instructions, which at least cover: translation and steering control of the mobile chassis, incremental displacement control of the two arms in three-dimensional Cartesian space, and opening and closing control of the end gripper.

[0012] This invention also provides a brain-computer interface execution method for continuous whole-body control of a humanoid robot, the method comprising the following steps: S1. Collect user's EEG data and perform online sliding window segmentation with a fixed time step to generate a continuous EEG data window; S2. Perform feature extraction and classification on each of the EEG data windows to obtain the original predicted category and the corresponding classification score; S3. Perform dual-threshold smoothing processing on the original prediction category, including: if the prediction confidence of the current time step is lower than a preset threshold, generate a "no action" instruction; if the prediction category of the candidate instruction is different from the current stable instruction, it needs to appear continuously for a preset number of confirmation steps before the candidate instruction is output as a safe execution instruction, otherwise generate a "no action" instruction. S4. Map the output safety execution command to the robot's incremental motion primitives to drive continuous actions, or map the "no action" command to the command to control the robot to maintain its current state.

[0013] Furthermore, in step S3, the dual-threshold smoothing process is performed, and the smoothing execution logic includes: setting... This indicates the current stable instructions maintained by the smoother. Indicates the decoding time step The actual safety instructions issued to the robot for execution ( (Indicates no action); the smoothing module actually executes instructions. The complete logical definition is as follows: in, Indicates the current decoding time step; Indicates the first Each decoding time step actually sends a safety instruction to the robot for execution; Indicates decoding time step Candidate instructions; Indicates decoding time step The prediction confidence level; This represents the preset confidence threshold. This indicates a stable instruction that has been confirmed in the previous decoding time step; This indicates a "no action" command; Indicates candidate instructions The number of consecutive occurrences in consecutive decoding time steps; This represents the minimum number of consecutive occurrences required to confirm a candidate instruction as a stable instruction; Indicates a logical "AND" relationship. This indicates other situations besides the conditions mentioned above. When the prediction confidence is lower than the preset confidence threshold, the smoothing module outputs a "no action" instruction; when the current candidate instruction is consistent with the previous stable instruction and the prediction confidence meets the threshold, the previous stable instruction is continuously output; when the current candidate instruction is inconsistent with the previous stable instruction, the candidate instruction is only confirmed as a new stable instruction and output as a safe execution instruction after the number of consecutive occurrences of the candidate instruction reaches the preset dwell confirmation step number; otherwise, a "no action" instruction is output to avoid the robot executing fuzzy or unstable control instructions.

[0014] Furthermore, the step of mapping the safe execution instructions to incremental motion primitives includes: Map chassis control commands to displacement increments with fixed steps; The control commands of the robotic arm are mapped into the displacement increment of the end effector in three-dimensional Cartesian space, and then calculated into joint angles through inverse kinematics. Map the gripper control commands to open and closed states.

[0015] The beneficial effects of this invention are as follows: Eliminating command jitter and ensuring physical safety: An innovative dual-threshold smoothing mechanism effectively identifies and filters invalid EEG signals generated by user gaze shifts, blinking, etc. Through dual protection of confidence rejection and dwell time confirmation, the robot's execution jitter rate is significantly reduced, preventing robot twitching or loss of control and ensuring the safety of human-computer physical interaction.

[0016] Balancing low latency and high stability, the sliding window step size and dwell confirmation cycle have been optimized. While suppressing decoding noise, the response latency of the control system is kept within the optimal range acceptable for human-computer interaction, achieving a smooth and continuous visual feedback loop, making the brain-controlled experience close to that of a conventional physical joystick.

[0017] It supports highly flexible and precise control. It abandons the traditional trigger-based fixed trajectory method and adopts a mapping strategy from commands to incremental motion primitives. Supported by 20 unified command spaces, users can use the brain-computer interface to safely and precisely manipulate the entire humanoid robot (chassis, arms, dexterous hands) like a joystick, completing complex grasping, placement, and movement tasks.

[0018] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall architecture of the brain-computer interface closed-loop system in an embodiment of the present invention. Detailed Implementation

[0020] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0021] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0022] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0023] See Figure 1This invention provides a brain-computer interface system for continuous full-body control of a humanoid robot. The system comprises four core modules connected in series and with information flow relationships: an EEG data stream acquisition and online sliding window segmentation module receives user EEG signals and outputs a continuous window; an EEG feature extraction and classification module decodes the window data and outputs the original prediction; a dual-threshold smoothing module filters and confirms the original prediction and outputs safety instructions; and an instruction space mapping and execution module converts the safety instructions into incremental motion primitives that drive the robot.

[0024] The system proposed in this invention breaks away from the traditional trial-based discrete control paradigm and is specifically designed for devices with high degrees of freedom and fine-grained continuous control requirements (such as full-body humanoid robots). It uses dual-threshold smoothing to convert the original model predictions into safe and stable robot execution instructions. The system as a whole consists of four tightly coupled core operating modules.

[0025] (1) EEG data stream acquisition and online sliding window segmentation In this embodiment, to meet the requirements of real-time closed-loop control, this module designs an online sliding window protocol for processing streaming SSVEP electroencephalogram (EEG) signals. The system sets the update interval for closed-loop control to... In this embodiment, The interval is 1 second, corresponding to a 4 Hz instruction inference frequency. It refers to the time interval between two consecutive EEG window extractions, feature decodings, and instruction updates performed by the system.

[0026] set up Indicates the first Each decoding time step. The decoding time step refers to the system operating according to the update interval. The time index for performing window extraction, feature decoding, and instruction update on the convection EEG signal is used. At each decoding time step... The system extracts data of length from the streaming EEG signals prior to the current moment. The latest EEG fragments form an EEG data window ,in, Indicates the number of brainwave channels. This indicates the duration of a single EEG data window.

[0027] Subsequently, the sliding window moves at an interval equal to the update interval. The step size advances forward, enabling the system to advance every [time / period]. Based on the current EEG data window A prediction result is generated independently, thus avoiding the cumulative delay caused by time aggregation and achieving a balance between real-time responsiveness and computational efficiency in continuous control.

[0028] (2) EEG Feature Extraction and Classification Module This module, serving as the SSVEP signal feature extraction and classification module, receives the first... EEG data window corresponding to each decoding time step SSVEP feature extraction is performed on it, and the output is targeted at... Original classification scores for each control instruction category .in, This indicates the total number of predefined control instruction categories. Indicates the first The first decoding time step The classification score corresponding to each control instruction category.

[0029] The original prediction category By selecting the original classification score The category corresponding to the highest score is obtained, and its calculation method is as follows: in, Indicates the first The original prediction category of the decoding model at each decoding time step, for robot control, also represents the candidate instruction initially obtained by the decoding model at the current decoding time step. Original prediction category Belongs to the predefined A set of control command categories is defined, and each category corresponds to a candidate control command for the robot through a preset category-command mapping relationship. This candidate command category is not directly issued to the robot for execution, but is instead input into a subsequent dual-threshold smoothing module. Only after confidence rejection and dwell time confirmation is it determined whether to output it as a safe execution command.

[0030] The decoding model in this embodiment is not limited to a specific EEG model architecture; any model capable of receiving EEG data can be used. Models that extract SSVEP EEG signal features and output multi-class prediction scores can all be applied to the system described in this invention, such as convolutional neural networks (CNN), Transformers, or traditional machine learning classifiers.

[0031] (3) Control-oriented dual-threshold smoothing strategy In continuous control, the EEG segments generated by the user during instruction switching often contain mixed or unstable responses. Furthermore, when the user's gaze is removed from the flashing target or when they blink, the generated EEG does not correspond to any intention. This module is designed with a dual-threshold smoothing mechanism to transform the unstable original prediction into a safe flow of execution instructions, and to explicitly output "no action" when the signal is invalid.

[0032] Threshold 1 is the confidence level for rejection. For the... Each decoding time step, the EEG feature extraction and classification module outputs the raw classification score. The dual-threshold smoothing module applies the original classification score. Perform Softmax normalization, and take the maximum probability value after normalization as the prediction confidence of the current decoding time step. The calculation method is as follows: in, Indicates the first Prediction confidence at each decoding time step; Indicates the first The probability values ​​of each control instruction category after Softmax normalization; Indicates the total number of control command categories.

[0033] like If the confidence level of the EEG signal at the current decoding time step is insufficient, the current candidate instruction is rejected. And output the "no action" command; among them, This indicates a preset reliability threshold. The preset reliability threshold... The system can be pre-set based on calibration or verification data. Specifically, it involves collecting calibrated EEG data from the user under each control command category, using a decoding model to output classification scores and calculate corresponding prediction confidence levels, and selecting thresholds based on the false trigger rate and correct recognition rate on the verification data. In this embodiment, The value is set between 0.6 and 0.9, preferably 0.8, to strike a balance between rejecting low-confidence invalid signals and retaining valid control commands.

[0034] Threshold 2 is a handover confirmation based on dwell time. Even if the prediction confidence level meets... Even within a single decoding time step, candidate instructions can still be affected by transient noise and undergo abrupt changes. Therefore, when the current candidate instruction... Compared with the previous stabilization command At the same time, the system does not immediately switch to a stable instruction, but instead counts the number of consecutive occurrences of the candidate instruction. Only when the candidate instruction is... The number of consecutive occurrences in the continuous decoding time step reaches the preset number of dwell confirmation steps. Only when the condition is met is the candidate instruction confirmed as a new stable instruction and output; otherwise, a "no action" instruction is output. This represents the minimum number of consecutive occurrences required to confirm a candidate instruction as a stable instruction. In this embodiment, This strikes a balance between acceptable control latency and the execution stability required for refined humanoid robot control.

[0035] The smooth execution logic is as follows: Let Indicates the first Each decoding time step actually sends a safety execution instruction to the robot for execution. Indicates the first The previous stable instruction has been confirmed for each decoding time step. Indicates the first Candidate instructions for each decoding time step, If the instruction indicates "no action", then the instruction will be executed safely. The method for determining it is as follows: in, Indicates the current decoding time step; Indicates the first Each decoding time step actually sends a safety instruction to the robot for execution; Indicates the first Candidate instructions for each decoding time step; Indicates decoding time step The prediction confidence level; This represents the preset confidence threshold. This indicates a stable instruction that has been confirmed in the previous decoding time step; This indicates a "no action" command; Indicates candidate instructions The number of consecutive occurrences in the decoding time step; This represents the minimum number of consecutive occurrences required to confirm a candidate instruction as a stable instruction; Indicates a logical "AND" relationship. This indicates any other situation besides the conditions mentioned above.

[0036] Confirming a candidate instruction as a stable instruction includes: when the prediction confidence is lower than the confidence threshold, the smoothing module outputs a no-action instruction; when the current prediction category is consistent with the previous stable instruction and the confidence meets the threshold, the previous stable instruction is continuously output; when the current prediction category is inconsistent with the previous stable instruction, the candidate instruction is only output as a new safe execution instruction after the number of consecutive occurrences of the candidate instruction reaches a preset number of resident confirmation steps, otherwise a no-action instruction is output to avoid executing ambiguous or unstable instructions.

[0037] Incremental motion mapping of instruction space To support full-body control of high-degree-of-freedom humanoid robots, this invention defines a unified instruction space covering core control dimensions. This unified instruction space includes... Class control instructions. Each safe execution instruction confirmed by the dual-threshold smoothing module. All of them are mapped to incremental motion primitives that can be executed at the robot's underlying level through a preset category-motion primitive mapping relationship.

[0038] Specifically, the The control commands include: 4 types of mobile chassis control commands, 12 types of dual-arm end effector control commands, and 4 types of end effector gripper control commands. The 4 types of mobile chassis control commands are for the robot to move forward, backward, turn left, and turn right, respectively mapped to the robot's forward / backward displacement and left / right turning in a planar coordinate system. The 12 types of dual-arm end effector control commands include for the left arm end effector to move up, down, left, right, forward, and backward, and for the right arm end effector to move up, down, left, right, forward, and backward, respectively mapped to fixed step displacement increments of the corresponding robotic arm end effectors in three-dimensional Cartesian space along the vertical, horizontal, and forward / backward directions, and solved for the corresponding robotic arm joint angles through inverse kinematics. The 4 types of end effector gripper control commands are for the opening and closing of the left and right grippers, respectively mapped to the opening and closing actions of the end effectors.

[0039] The chassis translation step size, chassis rotation angle, robotic arm end effector displacement step size, and gripper opening and closing range can all be preset according to the robot's structural dimensions, range of motion, and safety constraints. For continuous actions, the system executes safety commands continuously. The corresponding displacement or angle increment enables smooth and continuous movement of the robot chassis, arms, and end effector. For invalid commands or commands in the confirmation buffer period, the dual-threshold smoothing module outputs a "no action" command. At this point, the robot maintains its current state and does not perform any new actions, thereby avoiding erroneous actions caused by ambiguous EEG signals or unstable predictions, and ensuring the safety of human-computer physical interaction.

[0040] This invention is applicable to application platforms requiring precise, high-frequency continuous EEG control, and mainly includes: Rehabilitation medicine: Providing patients with severe motor dysfunction (such as ALS, high-level paraplegia) with real-time brain control capabilities for wheelchair navigation, exoskeleton assistance, or home service robots, greatly improving their autonomy in daily life.

[0041] Special hazardous environment operation: In high-risk environments such as deep-sea operations and nuclear radiation areas that are inaccessible to humans, operators can use this system to achieve safe, stable and precise remote operation of the remote bionic humanoid robot.

[0042] High-complexity human-machine collaboration scenarios: Supports brain-computer collaborative operation in intelligent manufacturing, controlling robotic arms and dexterous hands to complete complex and delicate object grasping and placement tasks.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A brain-computer interface system for whole-body continuous control of a humanoid robot, characterized by, The system includes: EEG data stream acquisition and online sliding window segmentation module: used to continuously segment streaming EEG signals at fixed time steps to generate a series of EEG data windows for real-time decoding; EEG feature extraction and classification module: used to receive the EEG data window, extract Steady-State Visual Evoked Potential (SSVEP) features, and output classification scores corresponding to each control command category at each decoding time step; wherein, the classification score is the score vector output by the decoding model for each control command category, the control command category corresponding to the highest score in the classification score is defined as the original predicted category, and the original predicted category is used as the candidate command for the current decoding time step; A dual-threshold smoothing module is used to convert the candidate instructions into safe execution instructions. The dual-threshold smoothing module includes: a confidence rejection unit, used to calculate the predicted confidence of the current decoding time step based on the classification score, and when the predicted confidence is lower than a preset confidence threshold, reject the current candidate instruction and output a "no action" instruction; and a dwell time confirmation unit, used to determine whether the current candidate instruction has appeared consecutively for a preset number of confirmation steps when it is different from the current stable instruction. If the preset number of confirmation steps is reached, the current candidate instruction is confirmed as a new stable instruction and output as a safe execution instruction; otherwise, a "no action" instruction is output. The instruction space mapping and execution module is used to map the safety execution instructions output by the dual-threshold smoothing module into incremental motion primitives of the robot to drive the robot to perform continuous actions; when a "no action" instruction is received, the robot is controlled to maintain the current state.

2. The whole-body continuous control brain-machine interface system for a humanoid robot according to claim 1, wherein In the EEG data stream acquisition and online sliding window segmentation module, the fixed time step is 0.25 seconds, and the EEG data window slides forward at the fixed time step to achieve a command inference frequency of 4Hz.

3. The whole-body continuous control brain-machine interface system for a humanoid robot according to claim 2, wherein In the EEG feature extraction and classification module, the prediction confidence is the maximum probability value of the classification score after normalization by the Softmax function; the confidence rejection unit compares the maximum probability value with the confidence threshold.

4. The whole-body continuous control brain-machine interface system for a humanoid robot according to claim 3, wherein The preset number of confirmation steps in the stay time confirmation unit is set to 3 steps.

5. The whole-body continuous control brain-machine interface system for a humanoid robot according to claim 4, wherein The instruction space mapping and execution module defines a unified instruction space containing 20 types of instructions, which at least cover: translation and steering control of the mobile chassis, incremental displacement control of the two arms in three-dimensional Cartesian space, and opening and closing control of the end gripper.

6. A brain-machine interface execution method for whole-body continuous control of a humanoid robot, characterized by, The method includes the following steps: S1. Collect user's EEG data and perform online sliding window segmentation with a fixed time step to generate a continuous EEG data window; S2. Perform feature extraction and classification on each of the EEG data windows to obtain the original predicted category and the corresponding classification score; S3. Perform dual-threshold smoothing processing on the original prediction category, including: if the prediction confidence of the current time step is lower than a preset threshold, generate a "no action" instruction; if the prediction category of the candidate instruction is different from the current stable instruction, it needs to appear continuously for a preset number of confirmation steps before the candidate instruction is output as a safe execution instruction, otherwise generate a "no action" instruction. S4. Map the output safety execution command to the robot's incremental motion primitive to drive continuous actions, or map the "no action" command to the command to control the robot to maintain the current state.

7. The brain-computer interface execution method for continuous whole-body control of a humanoid robot according to claim 6, characterized in that, In step S3, the dual-threshold smoothing process is performed to smooth the execution logic. comprising: providing a smoothing module at a current decoding time step the actual execution instructions are whose complete logical definition is as follows: wherein, denotes the current decoding time step; denotes the safety instruction actually issued to the robot for execution at the decoding time step; denotes the candidate instruction at the decoding time step ; denotes the prediction confidence at the decoding time step ; denotes the preset confidence threshold; denotes the stable instruction already confirmed at the previous decoding time step; denotes the "no action" instruction; denotes the number of consecutive occurrences of the candidate instruction in consecutive decoding steps; denotes the minimum number of consecutive occurrence steps required to confirm the candidate instruction as a stable instruction; denotes a logical "and" relationship, denotes other conditions in addition to the above conditions.

8. The brain-computer interface execution method for continuous whole-body control of a humanoid robot according to claim 7, characterized in that, The step of mapping secure execution instructions to incremental motion primitives includes: Map chassis control commands to displacement increments with fixed steps; The control commands of the robotic arm are mapped into the displacement increment of the end effector in three-dimensional Cartesian space, and then calculated into joint angles through inverse kinematics. Map the gripper control commands to open and closed states.

Images