Olfactory navigation rat robot based on brain-computer interface technology and navigation method

Abstract

The invention discloses an olfactory navigation rat robot based on a brain-computer interface technology and a navigation method. The olfactory navigation rat robot comprises a biological rat body, a signal acquisition stimulation backpack and a host, an invasive cortex array electrode is implanted in a rat olfactory bulb area, a field potential signal is accurately recorded, and motion control and respiration signal recording are achieved through the implanted electrode; the signal acquisition stimulation knapsack acquires electrophysiological signals of the rat in the movement process, and an online odor classification model is constructed by combining acquired respiratory signals or a neural network respiratory signal fitting model; based on a multi-normal form olfactory bulb signal database of rat individuals, a transfer learning technology is adopted to adapt to odor decoding tasks under different experimental conditions. The natural olfactory capability of the rat is utilized to construct an autologous neural signal closed-loop system, so that the accurate control of efficient site traversal and specific smell search on the rat robot is realized, and the accuracy and adaptability of olfactory navigation tasks are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of novel animal robots and control technologies, and in particular to an olfactory navigation rat robot and a navigation method based on brain-computer interface technology. Background Art

[0002] Animal robots are a new type of robot that uses animals as control objects. They are mainly based on brain-computer interface technology, which stimulates the relevant neural areas of the animal brain in a specific way to control the animal to complete specific movements in the way expected by the experimenter. At present, the main research content of animal robots includes the effects of different stimulation sites and how to use external sensors to better guide animal robots to complete target navigation. These studies have deeply explored the basic control mechanism of animal robots and built autonomous control processes around the movement characteristics of animal robots.

[0003] For example, a Chinese patent document with publication number CN107422352A discloses an animal robot autonomous navigation system and method, the system includes a GPS module, a multi-channel neural signal stimulation module and a control module arranged on the controlled animal, the GPS module is configured to receive GPS satellite signals and output the current position information of the animal robot to the control module; the control module is configured to receive the current position information of the GPS module, and compare it with the set target position information, determine the navigation route, and form a control instruction, the multi-channel neural signal stimulation module is configured to include a mirror current source, a boost circuit and an analog multi-way switch, a constant current is obtained by connecting the mirror current source and the boost circuit, and the corresponding channel of the analog multi-way switch is selected according to the control instruction to output the constant current to act on the neural control area of ​​the controlled animal, so as to realize the output of simulated neural coding information and realize navigation.

[0004] However, existing animal robots have not fully developed their advantageous performance and have not fully utilized the sensory capabilities of the controlled animals, such as the unique olfactory perception capabilities of rodents, in the process of building closed-loop control.

[0005] Many studies have shown that rodents have a strong sense of smell, and multiple sites in their olfactory neural circuits will produce unique responses to different concentrations and types of odors. Among them, the sense of smell of rats contains at least 1,000 olfactory receptors, which can distinguish at least tens of thousands of odors, and can distinguish at least 10 -10 mol / L of weak gas concentration. With further research in neuroscience, it was found that the decoding results of signals in multiple brain regions of rats are stable within a certain period of time. Compared with existing odor sensors, the olfactory perception unit of rats can perceive a richer variety and concentration of odors, and brain-computer interface technology can capture the response characteristics of this nerve.

[0006] Therefore, compared with the traditional animal robot system built with external sensors, the self-closed-loop olfactory navigation rat robot built with proprioceptive olfaction as the perception unit is more in line with the animal instincts of rats, and is more conducive to building a stable mapping relationship from external environmental information to internal neural responses, and can better tap the potential of animal robots. Summary of the invention

[0007] The present invention provides an olfactory navigation rat robot and a navigation method based on brain-computer interface technology, which utilizes the natural olfactory ability of rats to construct an autologous neural signal closed-loop system, thereby achieving precise control of the rat robot for efficient site traversal and specific odor search, and significantly improving the accuracy and adaptability of olfactory navigation tasks.

[0008] An olfactory navigation rat robot based on brain-computer interface technology, comprising a biological rat body, a signal acquisition stimulation backpack and a host;

[0009] The biological rat body is implanted with a small-diameter nasal tube in its respiratory tract, a cortical array electrode is implanted in its olfactory bulb area, and a pair of neural stimulation electrodes are implanted in the left and right areas of its medial forebrain bundle;

[0010] The signal acquisition stimulation backpack includes an inertial measurement unit, a respiratory pressure sensor, a neural signal recording module and a stimulation module;

[0011] The inertial measurement unit is used to collect the motion posture data of the biological rat body; the respiratory pressure sensor is connected to the nasal tube to collect the respiratory signal data of the biological rat body; the neural signal recording module is connected to the cortical array electrode to record the EEG signal data of the biological rat body during the breathing process; the stimulation module is connected to the neural stimulation electrode to stimulate the medial forebrain bundle site of the rat by sending stimulation instructions to the neural stimulation electrode;

[0012] The host is provided with a data processing unit. The host receives the motion posture, breathing signal and electroencephalogram signal data from the signal acquisition stimulation backpack through wireless communication technology. After processing by the data processing unit, the motion state of the rat and the current olfactory smell are judged, and different stimulation strategies are output according to the judgment results and sent to the stimulation module to guide the olfactory navigation rat robot to realize the site environment traversal and target odor search.

[0013] Preferably, a nasal tube with a diameter of 0.5 to 0.7 mm is implanted in one nasal cavity of the biological rat body.

[0014] Furthermore, the data operation unit aligns the received EEG signal data with the respiratory signal data, performs feature extraction and screening in the aligned time dimension, establishes and builds a multi-paradigm database and an online odor decoding model for the olfactory rat robot; during the actual navigation process, the real-time EEG signal and respiratory signal are input into the online odor decoding model to obtain the odor classification result under a single breath condition.

[0015] Furthermore, a navigation method of a rat robot based on olfactory navigation is characterized by comprising the following steps:

[0016] Step S0, preparation stage;

[0017] Connect the signal collection stimulation backpack to the biological rat body and complete the communication connection with the host computer;

[0018] Step S1, calibration phase;

[0019] The olfactory navigation rat robot is placed in a closed air cavity, and a plurality of corresponding odors of different types and concentrations are respectively input into the air cavity, and the EEG signals and respiratory signals collected by the signal acquisition stimulation backpack are transmitted to the data processing unit, which recalibrates the signals of the plurality of labeled odors;

[0020] Step S2, experimental preparation stage

[0021] Record the rat breathing curve physically or construct the rat breathing curve by neural network fitting; if the rat breathing curve needs to be fitted by neural network, jump to step S3;

[0022] Step S3, respiratory signal fitting stage;

[0023] The respiratory signal fitting model adopts an encoder-decoder architecture, inputs EEG signals of a specific frequency band, extracts the features of the input EEG signals with a three-layer convolutional neural network and a residual connection network, decodes the features with a long short-term memory network and an attention mechanism, and outputs a fitting curve of the respiratory signal or a classification result that identifies whether the data segment is in the exhalation phase or the inhalation phase.

[0024] Step S4, model processing stage;

[0025] First, the EEG signal data and the respiratory signal data are aligned, and the joint features in the time domain, frequency domain, and time-frequency domain in the inhalation segment are calculated. The feature set is constructed and input into the online odor decoding model to obtain the odor and concentration decoding results of a single breath. At the same time, according to the time domain features calculated by the inertial measurement unit, the motion posture classification model is input to obtain the current motion posture of the rat.

[0026] Step S5, navigation control stage;

[0027] According to the results of the online odor decoding model and the motion posture classification model, the rat's motion state and current olfactory odor are judged, and different stimulation strategies are output to guide the olfactory navigation rat robot to traverse the site environment and search for the target odor.

[0028] Furthermore, the online odor decoding model includes three models. The first model is a stationarity detection model, which is used to detect the stationarity of the input signal and eliminate the signal interference introduced by motion noise or contact noise. When the detection signal is stationary, it jumps to the second model; when the detection signal is non-stationary, it jumps to the third model.

[0029] The second model is an odor classification model, which extracts the time domain and frequency domain features of the input EEG signal and classifies and outputs the confidence probability of the odor type and concentration level;

[0030] The third model is the calibration model. If the decoding model output is available at the previous moment, the decoding data of the previous moment and the current decoding result are input into the calibration model at the same time, and the output of the decoding result is smoothed by Kalman filtering to reduce the possibility of single classification anomalies. If the odor classification model is skipped at the previous moment, the decoding data of the previous moment is further iterated in this model to output the decoding result.

[0031] Furthermore, the training of the online odor decoding model is based on a multi-paradigm olfactory bulb signal database of individual rats, using transfer learning technology to adapt to odor decoding tasks under different experimental conditions.

[0032] Optionally, in the model processing stage, the respiratory signal data is a respiratory signal collected by a signal collection stimulation backpack or a rat respiratory curve fitted by a neural network.

[0033] Furthermore, in the navigation control stage, the exploration process of the olfactory navigation rat robot is divided into an activation stage, a navigation stage and an abnormal stage using a finite state machine model;

[0034] During the activation stage, the olfactory navigation rat robot completes site traversal and environmental sniffing under the guidance of the anthropomorphic stimulation command sequence; when the odor online decoding model detects the target odor, it will switch to the navigation stage, and the stimulation command frequency will be linked to the target odor concentration gradient. Only when the target odor is detected will the corresponding stimulation command be given. The higher the target odor concentration gradient, the higher the stimulation command frequency, until the olfactory navigation rat robot is guided to approach the target odor source and complete the specific odor tracking task; in addition, the inertial measurement unit will monitor the motion state of the olfactory navigation rat robot in real time. Once the rat shows abnormal behavior, the model enters the abnormal stage and immediately stops the current stimulation.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The present invention constructs a bidirectional, self-closed-loop olfactory navigation rat robot based on brain-computer interface technology, which not only includes cortical array electrodes for recording neural signals, but also includes bipolar electrodes for applying bidirectional balanced electrical stimulation, as well as the body's breathing signals as marking information of neural signals. It also includes an inertial measurement unit for real-time recording of the motion state of the olfactory navigation rat robot, and has the ability to build closed-loop detection in neural signals and behavior.

[0037] 2. The olfactory navigation rat robot established by the present invention does not rely on external sensors as the input of characteristic signals. It uses its own perception as the signal source and brain-computer interface technology as a means to further give play to the special perception ability advantages of animal robots. At the same time, it also makes use of the unique movement ability of the original animal robot to realize a living, multi-type, movable olfactory sensor with certain active exploration capabilities.

[0038] 3. In the navigation method established by the present invention, the problem of recording breathing signals under actual conditions is fully considered, and a breathing signal fitting method is proposed, which can be used to solve the problem that pressure sensors cannot be used due to physical limitations and other factors.

[0039] 4. In the navigation method established by the present invention, the problems of neural signals recorded under actual conditions and decoder stability are fully taken into consideration, and a multi-odor online decoding model constructed by three sub-models is proposed. Different sub-models are adapted to the odor classification anomalies caused by abnormal input signals or abnormal classifiers, respectively, further enhancing the accuracy and stability of the odor classification task.

[0040] 5. The model migration method proposed in the present invention is adapted to the respiratory signal fitting model and multi-odor online decoding model provided in the present invention, and is used to further migrate the model trained on the training data set to the calibration data set to solve the problems caused by neural signal drift.

[0041] 6. The navigation method proposed in the present invention fully considers the activity habits of the navigation rat robot and uses a finite state machine model to control the rat robot to complete the odor search task in the venue. It is expected that this device can be extended to the search for drugs, explosives and objects with special odors in disaster relief environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of a signal acquisition stimulation backpack in the present invention;

[0043] Figure 2 It is a schematic diagram of a biological rat body wearing a signal collection stimulation backpack in the present invention;

[0044] Figure 3This is a data flow diagram of the olfactory navigation rat robot of the present invention;

[0045] Figure 4 This is a schematic diagram of the olfactory navigation rat robot of the present invention being located in a field for conducting experiments;

[0046] Figure 5 This is a flow chart of the navigation method of the olfactory navigation rat robot in the present invention. DETAILED DESCRIPTION

[0047] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0048] like Figures 1 to 3 As shown, an olfactory navigation rat robot based on brain-computer interface technology includes a biological rat body 1, a signal acquisition stimulation backpack 2 and a host 3.

[0049] A small-diameter nasal tube is implanted in the respiratory tract of the biological rat 1, a cortical array electrode is implanted in the olfactory bulb area, and a pair of neural stimulation electrodes are implanted in the left and right areas of the medial forebrain bundle. The small-diameter nasal tube is implanted in the right nasal cavity of the rat to record the unilateral breathing condition of the rat. The cortical array electrode is used to achieve the coverage area of ​​the olfactory bulb and record the neural signal response during the rat's breathing process. The double-stranded stimulation electrode is used to achieve precise stimulation of the medial forebrain bundle site of the rat robot, and control the rat robot to complete precise movement.

[0050] The signal acquisition stimulation backpack 2 includes an inertial measurement unit, a respiratory pressure sensor, a neural signal recording module, a stimulation module and a communication module. The inertial measurement unit is used to collect the motion posture data of the biological rat body 1; the respiratory pressure sensor is connected to the nasal tube to collect the respiratory signal data of the biological rat body 1; the neural signal recording module is connected to the cortical array electrode to record the EEG signal data of the biological rat body 1; the stimulation module is connected to the neural stimulation electrode to stimulate the medial forebrain bundle site of the rat by sending stimulation instructions to the neural stimulation electrode.

[0051] The host 3 includes a UI display unit, a data processing unit and a communication module. The communication module is responsible for receiving the return data from the signal acquisition stimulation backpack 2 using wireless communication technology. The data operation unit aligns the returned EEG signal with the respiratory signal, extracts and screens features in the aligned time dimension, and establishes and builds a multi-paradigm database and an online odor decoding model for the olfactory rat robot. In actual use, real-time data is input into the online odor decoding model to obtain the classification results under single-breath conditions.

[0052] In the embodiment of the present invention, the electrode disk of the implanted cortical array electrode is a 2*8 layout, which can cover an olfactory bulb area of ​​about 300*500um, and after implantation, the LFP signal of the right olfactory bulb area is recorded at a sampling rate of 2KHz.

[0053] The implanted bipolar stimulation electrode is made of two 50um nickel-chromium alloy wires twisted together. The double-strand stimulation electrode is used to achieve precise stimulation of the medial forebrain bundle site of the rat robot, guiding the rat robot to complete the olfactory exploration task.

[0054] The implanted nasal tube for recording breathing is a hollow tube with a diameter of 0.6 mm and a capped long needle inserted into the hollow tube. The hollow tube is made of stainless steel or other inert materials, with a length of 1 cm. The upper 1 / 3 of the hollow tube has a threaded structure that can be screwed with the capped long needle inserted into the hollow tube. The capped long needle is kept in the rat's head under non-experimental conditions to protect the tube from being blocked.

[0055] After considering the balance between data transmission rate and transmission power consumption during the communication process, the communication module selects the Bluetooth protocol as the communication protocol between the signal acquisition stimulation backpack 2 and the host 3. Its uplink path mainly transmits multiple sensors to collect and record data, and its downlink path mainly transmits control instructions issued by the host.

[0056] The detachable breathing pressure sensor consists of a detachable gas pipeline and a gas pressure sensor. During the actual recording process, one end of the gas pipeline is connected to the nasal tube for recording breathing implanted in the nose of the olfactory navigation rat robot, and the other end is connected to the gas pressure sensor. The gas pressure sensor detects the pressure change in the gas pipeline and converts its air pressure change into the change of electrical signal.

[0057] The neural signal recording module is built with a dedicated neural signal acquisition chip RHD2132. The front-end input is connected to the electrode interface, and the chip completes the analog-to-digital conversion of the neural signal.

[0058] The stimulation module is built with operational amplifier circuits to provide bidirectional balanced electrical stimulation to the stimulation target.

[0059] An embodiment of the present invention further provides an olfactory navigation rat robot navigation method based on brain-computer interface technology, which specifically comprises the following steps:

[0060] Step S0, this is the preparation stage. Connect the signal acquisition stimulation backpack with the olfactory navigation rat robot, specifically, connect the EEG signal acquisition interface on the backpack with the cortical array electrode recording electrode of the olfactory navigation rat robot, connect the stimulation interface on the backpack with the stimulation interface of the olfactory navigation rat robot, connect the hose of the collection respiratory sensor with the hard tube implanted in the nose of the olfactory navigation rat robot; turn on the electric control switch of the signal acquisition stimulation backpack, complete the signal connection with the host, and align the electrophysiological data with the respiratory data for transmission.

[0061] Step S1, this is the calibration stage. Place the olfactory navigation rat robot in an air cavity with a length, width and height of 50*30*30cm respectively, input the corresponding multiple marked odors into the air cavity respectively, and perform multi-odor signal calibration. The time for each odor input is 3 minutes. During the odor interval, pure air after filtering moisture is input into the air cavity as the signal baseline. The collected signal is also transmitted to the remote computing unit through the wireless communication module of the backpack, which recalibrates the signals of the multiple marked odors.

[0062] Step S2, this is the experimental preparation stage. According to the experimental site and actual needs, choose to record the rat breathing curve physically or fit it with a neural network. If you need to fit the rat breathing curve with a neural network, jump to step S3.

[0063] Step S3: Respiratory signal fitting stage. Considering the physical limitations of using gas pressure sensors to record the respiratory status of rats in the free movement paradigm, this step provides a respiratory signal fitting method for olfactory navigation rat robots based on brain-computer interface technology. The model adopts an encoder-decoder architecture, which can input neural signals of a specific frequency band, output a fitting curve of the respiratory signal or identify the classification result of the data segment as the exhalation phase or the inhalation phase. The specific settings can be configured according to the experimental requirements.

[0064] Step S4: This is the model processing stage. When processing the data uploaded by the signal acquisition stimulation backpack, the computing platform first aligns the neural data with the respiratory data, calculates the joint features in the time domain, frequency domain, and time-frequency domain in the inhalation segment, constructs a feature set, inputs the odor online decoding model, and obtains the odor and concentration decoding results of a single breath. At the same time, according to the time domain features calculated by the inertial measurement unit, the motion posture classification model is input to obtain the current motion posture of the rat.

[0065] Step S5, this is the navigation control stage. The navigation control method judges the rat's motion state and the current olfactory odor according to the results of the odor online classification model and the motion posture classification model, outputs different stimulation strategies, and guides the olfactory navigation rat robot to achieve site environment traversal and target odor search.

[0066] Furthermore, the present invention constructs neural and physiological signal data sets under multiple paradigms for animals with different numbers. It mainly uses a signal acquisition stimulation backpack to realize the recording of animal proprioceptive olfactory nerve signals and respiratory signals under multiple paradigms, including passive multi-concentration odor sampling, Go / No Go paradigm, and passive odor multi-concentration sampling and active odor sampling paradigms under reward stimulation. If reward stimulation is required in the experimental paradigm, a specific reward stimulation sequence is applied according to the experimental paradigm.

[0067] Furthermore, the input signal of the online odor decoding model is a feature set of multi-channel neural signals. Taking the inhalation moment of the respiratory cycle as the start and the next exhalation moment as the end, the neural data of the first 300ms or all the inhalation moments are extracted, and a set of features including but not limited to mean, extreme value, peak-to-peak value, multi-band spectrum energy, entropy, time-frequency energy, etc. is extracted.

[0068] Furthermore, the online odor decoding model will be constructed from three sub-models, among which the first model is a stationarity detection model, which will be used to detect whether the input neural signal is stationary and eliminate signal interference introduced by motion noise or contact noise. The second model is an odor classification model, which will classify and output the type and concentration level of odor according to the characteristics of the input neural signal, but will not directly output the result, but the confidence probability of each classification indicator. When performing online classification tasks, since there may be single neural signal classification anomalies, etc., a third calibration model is introduced to input the decoding data of the previous moment and the current decoding result into the calibration model at the same time, and smooth the output of multiple decoding results by means of Kalman filtering, etc., to reduce the possibility of single classification anomalies.

[0069] Furthermore, the breathing signal fitting model adopts an encoder architecture of a three-layer two-dimensional convolutional model and a residual connection network, and a decoder architecture composed of a long short-term memory network and an attention module. The breathing signal fitting model will provide two outputs, one is to directly provide the fitting of the breathing curve, and the other will provide a classification result, that is, whether the segment is in the exhalation stage or the inhalation stage. The experimenter can set the model output as needed.

[0070] Furthermore, in the model calibration phase, the original odor classification model and respiratory signal fitting model trained on the training data set will be further migrated to the calibration data set. Therefore, on the original odor classification model and respiratory signal fitting model, a deep domain adaptation network is further designed to learn the feature differences between the source data and the target data, and the adversarial learning method is used to extract the feature commonality of the training data set and the calibration data set to complete the migration task from the training model to the calibration data.

[0071] Furthermore, the motion posture classification model takes the three-axis acceleration and three-axis angular velocity data of the inertial measurement unit during the rat's movement as input, extracts its time domain features, and constructs a support vector machine classification model to divide the rat's movement state into four categories: stationary, sniffing, movement, and abnormal.

[0072] Furthermore, the navigation method of this animal robot is constructed with a finite state machine model, which is mainly divided into two stages. In the activation stage, the navigation method will guide the olfactory navigation rat robot to complete the site traversal with sequence stimulation. When the multi-odor classification mapping model detects the target odor, the model switches to the traversal stage. The navigation method will implement odor stimulation of different frequencies according to the target odor concentration gradient until the olfactory navigation rat robot is guided to complete the target odor tracking task.

[0073] In order to verify the effect of the present invention, a specific implementation process is introduced below.

[0074] In the embodiment of the present invention, the signal acquisition stimulation backpack 2 uses the RHD2132 chip to complete the acquisition of neural signals, uses the LSM6DS3 to complete the acquisition of the three-axis acceleration and three-axis angular velocity of the rat's motion state, uses the respiratory pressure sensor GZP6869A to complete the acquisition of the rat's respiratory signal, and uses the analog-to-digital conversion and operational amplifier circuit to form a stimulation module. The main control unit of the signal acquisition stimulation backpack is nRF52840.

[0075] The surgical procedure of the olfactory navigation rat robot is as follows: First, a cortical array electrode is implanted in the olfactory bulb area of ​​the rat to record the LFP signal at the olfactory bulb of the rat; in addition, a pair of neural stimulation electrodes are implanted in the left and right areas of the medial forebrain bundle of the rat to achieve precise stimulation of the rat robot; finally, a small-diameter nasal tube is implanted in its respiratory tract, which is used to record the unilateral breathing status of the rat. By inserting the pins of the signal acquisition stimulation backpack into the interface of the cortical array electrode on the rat's head, fixing the stimulation connector to the stimulation interface on the rat's head in the form of a cable, connecting the hose of the collection breathing sensor to the hard tube implanted in the nose of the olfactory navigation rat robot, turning on the electric control switch of the signal acquisition stimulation backpack, and completing the signal connection with the host, the assembly of the olfactory navigation rat robot is completed.

[0076] When the olfactory navigation rat robot Figure 4 When placed in the field to complete the scent search task, Figure 5 Process execution navigation method.

[0077] First, connect the signal acquisition stimulation backpack to the olfactory navigation rat robot. Specifically, connect the EEG signal acquisition interface on the backpack to the cortical array electrode recording electrode of the olfactory navigation rat robot, connect the stimulation interface on the backpack to the stimulation interface of the olfactory navigation rat robot, connect the hose of the respiratory sensor to the hard tube implanted in the nose of the olfactory navigation rat robot, turn on the electric control switch of the signal acquisition stimulation backpack, complete the signal connection with the host, and align the electrophysiological data with the respiratory data and transmit them out.

[0078] Secondly, the olfactory navigation rat robot was placed in an air cavity with a length, width and height of 50*30*30cm, and the corresponding multiple marked odors were input into the air cavity for multi-odor signal calibration. Each odor was input for 3 minutes. During the interval between odors, pure air after water filtration was input into the air cavity as the signal baseline. The collected signal was also transmitted to the remote computing unit through the wireless communication module of the backpack, which recalibrated the signals of the multiple marked odors.

[0079] Subsequently, the rat breathing curve is constructed by physical recording or neural network fitting according to the experimental site and actual needs.

[0080] When the host is processing the data uploaded by the wireless multimodal physiological signal acquisition and stimulation backpack, it first aligns the neural data with the respiratory data, calculates the joint features in the time domain, frequency domain, and time-frequency domain in the inhalation segment, constructs a feature set, inputs the odor online classification model, and obtains the odor and concentration decoding results of a single breath. At the same time, according to the time domain features calculated by the inertial measurement unit, it inputs the motion posture classification model to obtain the current motion posture of the rat.

[0081] Finally, the navigation method of this animal robot is constructed with a finite state machine model, and the navigation task will be completed in the following process. The model divides the exploration process of the olfactory navigation rat robot into two main stages, namely the activation stage and the navigation stage. In the activation stage, the olfactory navigation rat robot will complete the site traversal and environmental sniffing under the guidance of the anthropomorphic stimulation instruction sequence. Among them, the anthropomorphic stimulation instruction will simulate the operating habits of the operator to set the stimulation sequence to guide the olfactory navigation rat robot to complete the corresponding movement. When the odor online decoding model detects the target odor, the model will switch to the navigation stage, and the stimulation instruction frequency will be linked to the target odor concentration gradient. Only when the target odor is detected will the corresponding stimulation instruction be given. The higher the target odor concentration gradient, the higher the stimulation instruction frequency, until the olfactory navigation rat robot is guided to approach the target odor source and complete the specific odor tracking task. In addition, the inertial sensor will monitor the movement state of the olfactory navigation rat robot in real time. Once the rat has abnormal behavior, the model will enter the abnormal stage and stop the current stimulation immediately.

[0082] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An olfactory navigation rat robot based on brain-computer interface technology, characterized in that: It includes biological rat body, signal acquisition stimulation backpack and host; The biological rat body is implanted with a small-diameter nasal tube in its respiratory tract, a cortical array electrode is implanted in its olfactory bulb area, and a pair of neural stimulation electrodes are implanted in the left and right areas of its medial forebrain bundle; The signal acquisition stimulation backpack includes an inertial measurement unit, a respiratory pressure sensor, a neural signal recording module and a stimulation module; The inertial measurement unit is used to collect the motion posture data of the biological rat body; the respiratory pressure sensor is connected to the nasal tube to collect the respiratory signal data of the biological rat body; the neural signal recording module is connected to the cortical array electrode to record the EEG signal data of the biological rat body during the breathing process; the stimulation module is connected to the neural stimulation electrode to stimulate the medial forebrain bundle site of the rat by sending stimulation instructions to the neural stimulation electrode; The host is provided with a data processing unit. The host receives the motion posture, breathing signal and electroencephalogram signal data from the signal acquisition stimulation backpack through wireless communication technology. After processing by the data processing unit, the motion state of the rat and the current olfactory smell are judged, and different stimulation strategies are output according to the judgment results and sent to the stimulation module to guide the olfactory navigation rat robot to realize the site environment traversal and target odor search.

2. The olfactory navigation rat robot based on brain-computer interface technology according to claim 1, characterized in that: A nasal tube with a diameter of 0.5-0.7 mm is implanted in one nasal cavity of the biological rat body.

3. The olfactory navigation rat robot based on brain-computer interface technology according to claim 1, characterized in that: The data operation unit aligns the received EEG signal data with the respiratory signal data, performs feature extraction and screening on the aligned time dimension, and establishes and builds a multi-paradigm database and an online odor decoding model for the olfactory rat robot; During the actual navigation process, the real-time EEG signals and breathing signals are input into the online odor decoding model to obtain the odor classification results under single-breath conditions.

4. The navigation method of the olfactory navigation rat robot according to any one of claims 1 to 3, characterized in that: The following steps are involved: Step S0, preparation stage; Connect the signal collection stimulation backpack to the biological rat body and complete the communication connection with the host computer; Step S1, calibration phase; The olfactory navigation rat robot is placed in a closed air cavity, and a plurality of corresponding odors of different types and concentrations are respectively input into the air cavity, and the EEG signals and respiratory signals collected by the signal acquisition stimulation backpack are transmitted to the data processing unit, which recalibrates the signals of the plurality of labeled odors; Step S2, experimental preparation stage Record the rat breathing curve physically or construct the rat breathing curve by neural network fitting; if the rat breathing curve needs to be fitted by neural network, jump to step S3; Step S3, respiratory signal fitting stage; The respiratory signal fitting model adopts an encoder-decoder architecture, inputs EEG signals of a specific frequency band, extracts the features of the input EEG signals with a three-layer convolutional neural network and a residual connection network, decodes the features with a long short-term memory network and an attention mechanism, and outputs a fitting curve of the respiratory signal or a classification result that identifies whether the data segment is in the exhalation phase or the inhalation phase. Step S4, model processing stage; First, the EEG signal data and the respiratory signal data are aligned, and the joint features in the time domain, frequency domain, and time-frequency domain in the inhalation segment are calculated. The feature set is constructed and input into the online odor decoding model to obtain the odor and concentration decoding results of a single breath. At the same time, according to the time domain features calculated by the inertial measurement unit, the motion posture classification model is input to obtain the current motion posture of the rat. Step S5, navigation control stage; According to the results of the online odor decoding model and the motion posture classification model, the rat's motion state and current olfactory odor are judged, and different stimulation strategies are output to guide the olfactory navigation rat robot to traverse the site environment and search for the target odor.

5. The navigation method of the olfactory navigation rat robot according to claim 4, characterized in that: The odor online decoding model includes three models. The first model is a stationarity detection model, which is used to detect the stationarity of the input signal and eliminate the signal interference introduced by motion noise or contact noise. When the detection signal is stationary, it jumps to the second model; when the detection signal is non-stationary, it jumps to the third model. The second model is an odor classification model, which extracts the time domain and frequency domain features of the input EEG signal and classifies and outputs the confidence probability of the odor type and concentration level; The third model is the calibration model. If the decoding model output is available at the previous moment, the decoding data at the previous moment and the current decoding result are simultaneously input into the calibration model, and the output of the decoding result is smoothed by Kalman filtering to reduce the possibility of single classification anomalies. If the odor classification model is skipped at the previous moment, the decoding data of the previous moment is further iterated in this model to output the decoding result of this time.

6. The navigation method of the olfactory navigation rat robot according to claim 4, characterized in that: The training of the online odor decoding model is based on a multi-paradigm olfactory bulb signal database of individual rats, and uses transfer learning technology to adapt to odor decoding tasks under different experimental conditions.

7. The navigation method of the olfactory navigation rat robot according to claim 4, characterized in that: In the model processing stage, the respiratory signal data is the respiratory signal collected by the signal acquisition stimulation backpack or the rat respiratory curve fitted by the neural network method.

8. The navigation method of the olfactory navigation rat robot according to claim 4, characterized in that: In the navigation control stage, the finite state machine model is used to divide the exploration process of the olfactory navigation rat robot into the activation stage, the navigation stage and the abnormal stage. During the activation stage, the olfactory navigation rat robot completes site traversal and environmental sniffing under the guidance of the anthropomorphic stimulation command sequence; when the odor online decoding model detects the target odor, it will switch to the navigation stage, and the stimulation command frequency will be linked to the target odor concentration gradient. Only when the target odor is detected will the corresponding stimulation command be given. The higher the target odor concentration gradient, the higher the stimulation command frequency, until the olfactory navigation rat robot is guided to approach the target odor source and complete the specific odor tracking task; in addition, the inertial measurement unit will monitor the motion state of the olfactory navigation rat robot in real time. Once the rat shows abnormal behavior, the model enters the abnormal stage and immediately stops the current stimulation.

Images