806 results about "Visual observation" patented technology

The invention relates to an indoor robot navigation method based on multi-modal feature fusion. The method comprises the following steps: constructing a semantic map based on visual observation environment information; the method comprises the following steps: acquiring an RGB image of an indoor scene object, converting the RGB image into point cloud data, and preprocessing the RGB image and the point cloud data; image multilayer semantic features and point cloud features in the RGB image and the point cloud data are extracted respectively, and initial fusion is carried out; performing weighted fusion on the multi-layer semantic features and the point cloud features of the image by adopting space-channel-cross-modal multi-attention dynamic cooperation; and predicting a long-term target in a map space from top to bottom based on the fused feature map and the semantic map, and performing navigation path planning based on the current position and the long-term target. Under low-cost hardware configuration, the robustness of environment perception, the real-time performance of decision response and the usability of system integration in a dynamic complex environment are comprehensively improved, and the method is particularly suitable for application scenes such as indoor service robots.

The present disclosure provides a system for generating motor control commands for a humanoid robot, comprising an alpha model with over 1 billion parameters that processes visual observations and language instructions at a first frequency to generate contextual embeddings, and a beta model operating at a higher second frequency. The beta model includes an embodiment-specific state encoder projecting robot state information into a shared embedding space, a diffusion transformer module generating denoised action sequences through iterative flow-matching that cross-attends to the alpha model's contextual embeddings, and an embodiment-specific action decoder converting denoised sequences into motor control commands. The beta model generates action chunks comprising future action sequences over a predetermined time horizon in a single inference step, with the complete system having less than 5 billion parameters.

A cross-modal cross-layer target association method and apparatus, an electronic device, and a medium are disclosed, belonging to the field of sensor data fusion. The method includes: acquiring and preprocessing visual data and radar data; transforming a position coordinate of a radar observation target in the preprocessed radar data into a corresponding pixel coordinate; acquiring first radar observation targets projectable into a target detection box in the visual data according to the pixel coordinate corresponding to the radar observation target; calculating a pixel distance between a pixel coordinate corresponding to each first radar observation target and the target detection box; associating the radar observation target in the radar data and a visual observation target in the visual data with a tracking target; screening out a first tracking target associated with the visual observation target and not associated with the radar observation target; calculating a relative distance between each first radar observation target and a vehicle; and determining a radar observation target with the minimum relative distance and pixel distance as an associated target of the corresponding first tracking target.

The embodiment of the invention discloses an agricultural vehicle positioning method and device, electronic equipment and a storage medium, and the method comprises the steps: determining the posture, position and speed of an agricultural vehicle under a geographic coordinate system according to inertial navigation data obtained by an inertial navigation system; establishing an initial observation equation according to a satellite observation equation corresponding to a global satellite positioning system and a visual observation equation corresponding to a visual sensor; when it is detected that satellite data collected by a global satellite positioning system is valid, the initial observation equation serves as a target observation equation; when it is detected that the satellite data collected by the global satellite positioning system is invalid, the initial observation equation with the satellite observation equations removed serves as a target observation equation; the current error state quantity of the inertial navigation system is corrected according to the target observation equation, and the attitude, position and speed of the agricultural vehicle are corrected according to the updated error state quantity, so that the actual positioning information of the agricultural vehicle is obtained, and the robustness of the positioning result of the agricultural vehicle is improved.

The invention discloses a security reinforcement learning-driven large language model security decision agent, and the decision agent comprises a high-level semantic planner which is used for receiving a target and constraint instruction in a text form, receiving a language or visual observation signal of an environment at the same time, and outputting text formatted security risk information and suggested action planning; the low-layer action actuator is used for receiving low-dimensional observation and semantic codes of the environment, and the semantic codes are output by the high-layer semantic planner after text embedding conversion; the strategy network of the low-layer action actuator outputs a final safety action; the training alignment module is used for optimizing the strategy network and the value network; a high-level semantic planner is fed back and prompted through reward and cost signals collected through environment interaction, and parameters of a strategy network and a value network are trained through a security reinforcement learning algorithm. According to the method, the decision cannot violate the given text security constraint while the decision of the given text target is completed.

The invention discloses a diffusion model and explicit constraint guidance fused visual navigation method and system. The method comprises the following steps: performing feature extraction on visual observation data to obtain target observation semantic features; performing feature extraction on the navigation target data to obtain target image features; performing feature fusion on the target observation semantic feature and the target image feature to generate a condition vector; inputting the condition vectors into a condition diffusion model for diffusion training to generate a plurality of initial candidate paths; constructing an explicit constraint based on the initial image data, and performing path optimization on each initial candidate path according to the explicit constraint to obtain a plurality of candidate optimization paths; the explicit constraints comprise collision constraints and target distance constraints; and determining a target navigation path from each candidate optimization path according to a path selection strategy. According to the method, the problems of insufficient path safety and poor dynamic adaptability can be solved, the safety and accessibility of the path are remarkably improved, the path navigation efficiency is further improved, and the method can be widely applied to the technical field of computers.

The invention provides a glass wine bottle surface defect detection system. The glass wine bottle surface defect detection system comprises the following steps: S1, carrying out image acquisition; s2, a step of image processing; s3, a step of defect identification and classification; the step S1 specifically comprises the following steps: S11: selecting and configuring equipment, selecting an industrial-grade high-definition camera which has the characteristics of high resolution, low noise and high dynamic range according to the size, the shape and the surface characteristics of the glass wine bottle so as to ensure that a clear image is captured, and configuring an LED backlight or annular light source system so as to eliminate shadow and light reflection, so that the glass wine bottle is protected; details on the surface of the wine bottle are highlighted, and brightness and color temperature of the light source are adjusted according to scenes; according to the method, the problems of low efficiency and low detection accuracy caused by experience judgment or manual visual observation during existing glass wine bottle detection are well solved, and the intelligent detection method based on visual observation is provided to improve the detection efficiency and accuracy.

The invention relates to a rotary micro-liquid membrane reactor in-situ detection system for monitoring the nucleation behavior of a nano material. The reactor comprises a rotor and a stator, the base angle of the stator is 70-85 degrees, the rotating speed of the stator is 500-5000 rpm, the width of a slit is 10-500 microns, it is guaranteed that fluid does not backmix when passing through the slit, and a strong shear field is formed. A micro-negative pressure sampler is manufactured to capture nucleation particles in a high-speed shear field to enter a visual observation area, and a high-speed camera and a microscope objective lens are connected in situ, so that an instantaneous image of a nucleation process is obtained in a nano-micro scale, and a transient evolution behavior of a crystal nucleus is captured in situ. The construction of the rotary micro-liquid membrane reactor in-situ detection system solves the problems that the large-scale preparation quality is poor due to the fact that a traditional stirred tank reactor is poor in structure and micro-mixing effect and uncontrollable in material nucleation and growth, and the instantaneous phase transformation and nucleation process cannot be observed due to the fact that the reactor is closed. The obtained hydrotalcite nucleation result enriches the non-classical nucleation theory, and provides guidance for large-scale preparation of nano materials.

The invention relates to the technical field of decorative materials, in particular to a decorative material with a 3D texture structure and a preparation method and application of the decorative material. The method comprises the following steps: carrying out digital printing on a pretreated substrate to obtain a 2D pattern layer on the surface of the substrate; coating the surface of the 2D pattern layer with a photocuring coating, carrying out preliminary curing, and then carrying out ink-jet printing of a texture developing solution on the surface subjected to preliminary curing to carry out secondary curing, so as to obtain a cured coating; the cured coating comprises a hardened part and an unhardened part; and after an unhardened part in the cured coating is removed through polishing and wire drawing, a surface coating is coated, and the decorative material with the 3D texture structure is obtained. The textures with different widths and depths can be obtained in a manner of combining primary curing of the coating with ultraviolet absorption of the developing solution, and then a microcosmic coarse and visual smooth form is obtained through polishing and wire drawing treatment, so that the obtained 3D texture structure is closer to natural textures.

The invention is suitable for the technical field of rock grouting reinforcement, and relates to a visual crossed fracture flowing water grouting test device and method. A transparent soil material with soil force properties and hydraulic properties similar to those of natural soil such as an internal friction angle, a deformation characteristic and a flow boundary effect is adopted; the water flow and slurry flowing state in an actual stratum can be truly simulated, and the accuracy of a test result is remarkably improved; by means of the layered water injection design of the soil filling box, comprehensive observation of the grouting behavior rule of cross fractures in stratums in different water saturation states is achieved, and the influence of complex hydrological conditions on the grouting effect is effectively reproduced; and a tracer agent with observation advantages is introduced in the grouting process, so that the visual observation capability on the slurry diffusion path, the slurry diffusion range and the flow velocity change is greatly enhanced. According to the method, the diffusion path, the expansion range and the grouting effect of the grout in a deep crack and a large model can be visually observed, the grouting effect is greatly improved, and a scientific basis is provided for optimizing the grouting process and predicting the grouting effect.

The invention provides a robot control method based on a visual language action model and related equipment thereof, and relates to the technical field of intelligent robots, and the method comprises the steps: obtaining visual observation information, a target language instruction and robot body state information of a target robot; reasoning the visual observation information, the target language instruction and the robot body state information through a pre-trained visual language action model to obtain a predicted low-dimensional potential vector; reconstructing the predicted low-dimensional potential vector into a robot action instruction through a pre-trained high-dimensional action reconstruction module; wherein the dimension of the robot action instruction is matched with the action space dimension of the target robot; and the target robot is driven to execute the high-dimensional robot action instruction. According to the method, the limitation of the existing visual language action model on the action output dimension can be solved, so that the high-degree-of-freedom robot is effectively controlled.

The invention provides a road inspection target positioning method based on multi-source data fusion, and relates to the field of road daily inspection and infrastructure performance monitoring, and the method comprises the following steps: obtaining multi-source data based on a preset multi-source sensor; through deep coupling of satellite positioning carrier double-difference observation and inertial navigation solution of an inertial sensor, error correction is carried out in combination with Kalman filtering recursion, and fusion navigation data is determined; according to the fused navigation data, carrying out fusion processing on the two-dimensional visual observation data and the three-dimensional space point cloud data to obtain a forward view depth map; determining the national geodetic coordinates of the disease by combining a depth value back calculation method and utilizing a forward view depth map; and obtaining a starting kilometer pile number in a road inspection process, and determining a mileage pile number of the disease by combining the national geodetic coordinate of the disease and utilizing a Gaussian positive algorithm. According to the invention, centimeter-level geographic coordinate positions of diseases and corresponding mileage stake numbers can be output.

The invention discloses a manifold perception-based large model navigation method guided by a historical topological graph, and relates to a computer vision technology. The method aims at solving the challenges that in the navigation process, long-distance reasoning experience is insufficient, instruction fragments and dynamic visual observation are difficult to align, and large model reasoning is prone to illusion interference. Firstly, a large model based on an encoder-decoder structure is used for supplementing historical information coding for a visual observation sequence, and therefore global topological information guidance is provided for long-distance reasoning. And secondly, in order to effectively solve the problem that large model reasoning is subjected to illusion interference, significant space-time differences in a visual observation sequence are mined by using a multi-curvature manifold, so that the large model can accurately describe the current environment and make a decision according to a visual reference object in thinking. Besides, in order to strengthen the perception capability of the large model to the space structure and establish a graph self-attention mechanism, the node distance embedded in the constructed historical topological graph is combined with the visual similarity so as to model the space relationship between the nodes.

The invention discloses a test device and a test method for filling fracture seepage-erosion visualization. The test device comprises a fracture overflowing model, a fluid flow control system and an image and pressure data acquisition and analysis system. The fracture overflowing model has the appearance of a real rock fracture surface, and the opening degree can be quantitatively adjusted through a filler gauge gasket; the fluid flow control system uses a peristaltic pump and a stirrer to provide constant-flow fluid and particle input for the fracture overflowing model; according to the test device, the light transmission principle is utilized, quantitative and visual observation and research of the seepage-erosion process of the filled fracture are achieved, and form and permeability evolution data of a particle erosion area in the filled fracture are accurately obtained.

The invention discloses a VLA model autonomous generalization method, system and device and a medium, and belongs to the technical field of robot control. The method comprises the steps of obtaining visual observation data and natural language task description of a current environment of a robot, generating an action sequence and controlling the robot to execute tasks; track data in the task execution process are collected, and token-level dense rewards and structured reflection information are output; storing the trajectory data, the token-level dense reward and the structured reflection information into a double-memory block; extracting task priori from the long-term memory by using an experience-driven generator, and generating trajectory data containing a sub-target chain; updating the VLA strategy model by using a reinforcement learning algorithm based on the token-level dense reward; and updating the VLA strategy model by using a supervised fine tuning algorithm based on the high-quality track data set. According to the method, the technical problem of autonomous generalization of the VLA model is solved.

The invention provides a variable environment visual language navigation method and system based on historical reflection, and relates to the technical field of computer vision, natural language processing and robot navigation. The navigation history information and the corresponding code embedding instruction used in the scene graph code embedding and next action are obtained; a gating network and a reflection network are utilized to analyze navigation historical information used in scene graph code embedding and next action and a corresponding code embedding instruction, and a correction instruction is obtained; the reflection network is a large language model used for generating a correction instruction; and based on the correction instruction, performing calculation by utilizing a graph perception self-attention mechanism, performing path navigation on the intelligent agent, obtaining a visual language navigation result, and completing visual language navigation in the variable environment. According to the method, the problems of poor adaptability and low fault-tolerant capability of visual language navigation are solved.

The invention discloses a micro-fluidic chip for basalt CO2 mineralization reaction and a preparation method, the micro-fluidic chip comprises a bottom glass plate, a middle epoxy resin layer and a top glass plate, the preparation process comprises the following steps: installing a metal injection / outpouring pipeline on the bottom glass plate drilled with corresponding bolt holes, and forming an accommodating cavity by using a sealing ring; preparing and screening basalt particles according to the particle size distribution of a target reservoir, and filling the accommodating cavity with the basalt particles; quantitatively analyzing a particle arrangement structure based on a digital image processing technology, ensuring that the geometric similarity between the particle arrangement structure and a target reservoir is greater than or equal to 90%, and if the particle arrangement structure does not reach the standard, refilling; after the verification is passed, covering a top glass plate and pressing by using a bolt; and finally, pouring epoxy resin and sealing to form the micro-fluidic chip. The chip can bear high temperature and high pressure, supports visual observation and quantitative analysis of CO2 mineralization reaction in a deep geological environment, has high-precision pore reproduction, long-period sealing performance and high preparation efficiency, and is suitable for interdisciplinary research in the fields of energy environment and geophysics.

The invention provides an indoor navigation method based on multi-modal data fusion and large model enhancement, which comprises the following steps: acquiring multi-modal data including Bluetooth signal intensity data, IMU (Inertial Measurement Unit) motion data and image data through a user mobile terminal, and constructing multi-source sensor input; iMM-EKF is adopted to carry out deep fusion on multi-modal data, different motion models are established, different motion states of a user can be adapted, and accurate navigation can be output; the accuracy and the anti-interference performance of the system are improved by performing optimization enhancement through a large model, compensating IMU (Inertial Measurement Unit) through MLP, enhancing visual observation through Transformer and dynamically estimating IMM-EKF noise parameters through LSTM (Long Short Term Memory); and regulating and controlling output weights of the IMM-EKF and the large model through an A2C reinforcement learning framework strategy, and outputting final positioning.

The invention provides an inertia / satellite / vision adaptive integrated navigation method and system. The method comprises the following steps: carrying out inertia measurement and navigation calculation to obtain inertia data; calculating satellite navigation observed quantity according to the satellite navigation data and the inertial data, and constructing a satellite observation model; calculating visual navigation observed quantity according to the visual navigation data and the inertial data, and constructing a visual observation model; aiming at a satellite and a visual observation model, respectively adopting an innovation-based adaptive covariance estimation method to obtain satellite and visual observation noise covariance updated values; constructing a satellite and visual navigation health degree, and calculating a satellite and visual fusion weight according to the satellite and visual navigation health degree; controlling updating of satellite and visual navigation observed quantity according to the satellite and visual fusion weight; and calculating a weighted equivalent observation matrix and a noise covariance, and carrying out filtering estimation. According to the method, an adaptive noise estimation and sensor health degree evaluation mechanism is introduced, dynamic weighted fusion of multi-sensor data is realized, and the robustness and precision of a navigation system in complex environments such as satellite signal lock losing and visual feature missing are improved.

The invention belongs to the technical field of artificial intelligence and robot autonomous navigation, and discloses an open vocabulary topology semantic graph construction and navigation method and system for an unknown environment. A topological map composed of nodes and edges is constructed in an online incremental mode, the nodes represent key positions in the environment, and the edges represent passable connection between the positions; for each node in the topological map, coding the corresponding visual observation image into a semantic feature vector by using a pre-trained visual language model; performing graph neural network fusion on the semantic feature vectors, executing cross-modal alignment with a natural language instruction, and generating global map representation; and based on the global map representation, generating a navigation action, and driving an intelligent agent to move towards the target. According to the method, an environment representation which deeply fuses spatial topology and semantic information is constructed, and intelligent decision making is carried out according to the environment representation.

The invention belongs to the technical field of oil and gas drilling well control equipment, and particularly relates to a ram blowout preventer state confirmation method with a locking indication window, which comprises the following steps: arranging a transparent observation window on a blowout preventer shell, and calibrating the alignment position of the end face of a locking shaft and an inner side reference scribed line; after the locking action is executed, the alignment state is confirmed through multi-angle visual observation; a switchable auxiliary light source is integrated to eliminate visual interference; an industrial camera is used for collecting images, displacement deviation is recognized through an edge detection algorithm, and an alarm is given if the displacement deviation exceeds the limit; and generating an encrypted digital report containing a timestamp and equipment information, uploading the encrypted digital report to a monitoring platform, and supporting real-time check and electronic signature of a mobile terminal. Through three layers of mechanisms of standardized vision-mechanical mapping, optical enhancement and digital perception, high-reliability, high-compatibility and traceable locking state confirmation is realized.

The invention discloses a bidirectional visible narrow rectangular flow channel CHF test device and test method, and belongs to the field of nuclear power. The bidirectional visible narrow rectangular flow channel CHF test device comprises a pressure-bearing module and a transparent heating module, the pressure-bearing module comprises a pressure-bearing frame, a rectangular flow channel is arranged in the pressure-bearing frame, and through observation windows are arranged on the two sides of the pressure-bearing frame; the transparent heating module comprises a transparent heating assembly and a power supply module, the transparent heating assembly is installed on the observation window in a sealed mode and defines a test space, the transparent heating assembly comprises an ITO transparent conducting film plated on an ITO transparent conducting film substrate, and the transparent heating assembly provides heat input and a visible observation area in the CHF test. The test device can provide bidirectional visible CHF test conditions and effectively improve the test observation precision.

The invention relates to the technical field of grouting tests, in particular to a three-dimensional high-pressure splitting grouting test method, which comprises the following steps: laying a rock mass sample in a large cubic test box, and applying a stable three-dimensional stress field to the sample in three directions by using a multi-hydraulic cylinder array; grout is injected in a stress maintaining state, grouting pressure, grouting flow and hydraulic cylinder data are synchronously collected to generate a pressure-flow-time relation curve, and meanwhile the grout diffusion process is directly observed through a visual window. According to the method, boundary interference is effectively reduced through a large-size sample and uniform loading, so that slurry is naturally diffused in a stress environment close to a real stratum. The dynamic stress compensation technology maintains boundary stability, and ensures that splitting expansion is not influenced by artificial constraint. A pressure-flow-time curve is combined with visual observation, and the diffusion rule of slurry in a three-dimensional stress field is clearly presented. According to the method, a test means closer to reality is provided for the underground engineering grouting process, and the reliability of grouting parameter design is remarkably improved.

The invention relates to the technical field of rock mass fissure medium seepage experiments, in particular to a layered filling fissure seepage-erosion visual experiment device which comprises a rack horizontally fixed to an experiment table; the plane light source is horizontally fixed on the rack, the adjustable brackets are distributed on two sides of the experiment table, and the layered filling crack models are arranged on the two adjustable brackets; the injector is arranged on the experiment table and is communicated with the fluid inlet through a pipeline; the liquid collecting bottle is arranged below the fluid outlet; the quality monitoring module, the pressure monitoring module and the image acquisition module are arranged on the experiment table. A layered filling fracture model is manufactured by filling a fracture model in sections, the layering characteristic of a soil body can be reflected, and real-time monitoring of filling fractures is achieved by connecting a high-precision spiral pull rod pressure sensor and constructing a quality monitoring system. And a pressure monitoring system and an image acquisition module are matched to realize visual observation of the filled fracture seepage-erosion experiment.

According to the pressure damage dynamic risk assessment method based on multi-modal sensing, multi-physical field data is collected in real time through an MEMS piezoresistive array, a near infrared spectrum sensor and a flexible capacitance type humidity sensor, and early precise early warning and automatic grading intervention are achieved in combination with an LSTM neural network and dynamic threshold adjustment. According to the method, the early warning time of the stress injury is advanced by 4-6 hours, so that the clinical misjudgment rate is reduced to be less than 15% from 38%, and the problems that in the prior art, poor blood circulation of subcutaneous tissues cannot be accurately identified by traditional visual inspection, the hysteresis of the currently adopted preventive measures and the clinical accurate judgment error rate are relatively high, and the clinical accuracy is relatively low are solved. In order to solve the problem that the occurrence risk and the compression degree of the stress injury cannot be accurately judged due to the fact that the existing tissue ischemia assessment method is based on the three-layer architecture of'multi-modal sensing and dynamic modeling combined with intelligent closed loop ', the transformation from'passive discovery' to'active prevention 'is realized, and a more comprehensive tissue ischemia assessment system is constructed.

The invention discloses a simulation fracture multiphase flow visualization test system of a variable structure in a dynamic water environment. The simulation fracture multiphase flow visualization test system comprises a simulation fracture model, a hydrodynamic force control module, a visualization observation module and a supporting structure. The simulation fracture model is composed of a transparent fracture structure and an external clamp, the transparent fracture is prepared by rolling over rock samples which are obtained through splitting or 3D printing and have different roughness and sizes, and the fracture opening degree and the filling degree are flexibly adjusted and controlled by adjusting the thickness of a gasket and the volume of filler. And the visualization module consists of a camera and a light source which are arranged above and below the crack model and is used for recording the multiphase flow process. The hydrodynamic control module is composed of a Markov bottle, a flow meter, an injection pump and a pressure sensor and is connected with the fracture model through a hose, and accurate control over the dynamic water environment and the second-phase liquid injection condition is achieved. The device has the characteristics of being variable in fracture rough structure and size, controllable in hydrodynamic condition, visualized in flowing process and the like, and provides important support for researching the fracture multiphase flowing law.

The embodiment of the invention relates to the technical field of robot moving trajectory planning, and provides a robot moving trajectory optimization control method and device, electronic equipment and a medium, and the method comprises the steps: obtaining visual observation data and a natural language task instruction when a robot executes a current moving trajectory; joint reasoning is carried out on the visual observation data and the natural language task instruction, a discrete action sequence is generated, smooth optimization is carried out, and a first continuous running track is generated; the robot is controlled to be converted from the current moving track to execute the first continuous moving track, and the visual observation data and the natural language task instruction of the robot executing the first continuous moving track continue to be obtained; and smooth transition processing is carried out on the newly generated second continuous moving track and the first continuous moving track, and the robot is controlled to be transited to the second continuous moving track from the first continuous moving track. Therefore, smooth trajectory generation meeting dynamic constraints, continuous trajectory transition under dynamic instructions and safety control in an abnormal state are realized.

The invention discloses a folding type clamping device, relates to the technical field of medical instruments, and is suitable for working channels of superfine endoscopes such as a breast duct endoscope. The clamping device comprises a clamping assembly, a transmission assembly, a limiting assembly and an operation assembly. The clamping assembly comprises a metal pipe body and an integrally-formed retractable net basket structure, the transmission assembly is a traction steel wire with the front end connected with the net basket structure, the limiting assembly limits the stroke of the traction steel wire through a bridge type metal pipe and a movable pin, and the operation assembly drives the traction steel wire to move. The mesh basket structure is in a closed state without external force and can be inserted into an endoscope channel; after the tissue leaves the channel, the operation assembly drives the traction steel wire to control the net basket to open and close to clamp the tissue in combination with visual observation. The device is made of medical stainless steel, is a disposable sterilized product and can be used after being disassembled. The device breaks through the sampling bottleneck of a superfine endoscope, sampling is stable and safe, early diagnosis of breast diseases is assisted, and the clinical value of the endoscope is improved.

The utility model relates to the field of batteries, and discloses a battery in-situ test sample cell, which comprises a first shell and a second shell, a first accommodating cavity with one closed end is arranged in the first shell, the first accommodating cavity is used for accommodating a sample to be tested, and the second shell covers the first shell so that the other end of the first accommodating cavity is closed. The top of the second shell is provided with an observation window, the observation window can observe the to-be-tested sample in the first containing cavity through the observation channel, an annular second containing cavity is formed outside the first containing cavity, and a semiconductor temperature control chip is installed in the second containing cavity. The device can be combined with synchrotron radiation analysis and gas product detection, has a lateral visual observation function, is reasonable in whole structural design and high in compatibility, can stably operate in a temperature range of-20 DEG C to 70 DEG C, and is adaptive to various in-situ characterization platforms.

The application discloses a safe decision-making agent of a large language model driven by safe reinforcement learning, and the decision-making agent comprises: a high-level semantic planner, which is used for receiving target and constraint instructions in a text form, receiving language or visual observation signals of an environment, and outputting safe risk information and a suggestion action plan in a text format; a low-level action executor, which is used for receiving low-dimensional observation and semantic encoding of the environment, wherein the semantic encoding is output by the high-level semantic planner after text embedding conversion; a strategy network of the low-level action executor, which outputs a final safe action; a training alignment module, which is used for optimizing the strategy network and a value network; reward and cost signals collected through environment interaction are fed back to the high-level semantic planner, and parameters of the strategy network and the value network are trained through a safe reinforcement learning algorithm. The application is convenient for realizing the decision-making of a given text target while ensuring that the decision-making does not violate a given text safety constraint.