Abstract: First, an intelligent fixed robot based on neural networks is designed. This framework can incorporate advanced algorithms to expand the system's functions. Then, a solution for collision avoidance intelligent decision-making system based on multi-control system is proposed to ensure safe use and prevent malfunctions. Finally, a framework structure based on multi-agent system is given to complete the overall design of the intelligent robot. Keywords: Home fixed robot; Intelligent decision-making; Multi-agent; Neural network 1 Introduction The development of home appliance technology is an important indicator and manifestation of a country's scientific and technological level and industrial automation. The application of home appliance automation is becoming increasingly widespread, but the development of home appliance automation is still in its initial stage and has not yet reached a stage of easy operation. In fact, it can be considered a stationary robot that operates through embedded software, perceives through sensors, and communicates with users through networks. This technology integrates computer science, cybernetics, mechanics, information and sensing technology, artificial intelligence, bionics, and other disciplines to form a high-tech innovation, integrating the development achievements of many disciplines. It represents the development of high technology and is a key aspect of scientific and technological research. With the rapid progress of computer and electronic information technology... The development of home appliance control technology is accelerating, and its intelligence is increasing. It is a comprehensive system integrating dynamic decision-making and planning, behavior control and execution, and other functions. It is a type of intelligent system that can adjust the state of home appliances by sensing the environment and its own state through sensors. The optimization and adjustment of the automatic target recognition framework algorithm must adopt a visual interface design, while providing opportunities for design selection and improvement. The system's interface consists of interconnected modules, each representing a separate subsystem. The framework should be able to optimize all or part of the program and should add as many new optimization algorithms as possible. The framework can provide online help for algorithm developers to familiarize themselves with the user interface and framework usage. The framework should fully utilize existing object-oriented programming design environments; design choices used for guidance can be embedded in inheritance structures, and the system should be easily extended by generating derived classes or adding interfaces. The framework adopts a modular design through relatively independent module design. According to the characteristics of object-oriented programming, most of the designer's work involves writing programs, deriving new objects from existing objects, and merging them together. This work is mainly accomplished using inheritance extension functions. 2. Automatic Target Recognition Framework Based on Simplified Structure Since this is the initial stage of the experiment, this paper proposes a hierarchical and modular automatic target-priority framework. Utilizing the principle of hierarchical recognition, a complex multi-category recognition problem is decomposed into simpler recognition problems at different levels within a multi-layered system. This allows the recognition system to flexibly employ corresponding feature extraction and target recognition algorithms at different levels. Furthermore, adding new recognition types only requires adjustments to the local system structure or parameters. In addition, a practical automatic target recognition system typically needs to possess basic functions such as data acquisition, feature extraction, and target recognition and classification. This type of home appliance system also suffers from drawbacks such as high system cost, numerous transmission lines hindering installation, and insufficient computer expansion, making it unsuitable for ordinary households. Adaptability to different tasks and special environments is a key difference between home appliances and general automated equipment. Intelligent home appliances have far surpassed the shapes and limitations of traditional home appliances in terms of appearance, better meeting the specific requirements of various home environments. Their significantly enhanced functionality and intelligence have opened up broader development space for home modernization technology, and multi-sensor information fusion technology is a key technology for intelligentization. (1) There are many types of sensors used by sensors. According to different uses, they are divided into two main categories: internal measurement sensors and external measurement sensors. Internal measurement sensors are used to detect the internal state of electrical components, including: temperature sensors. External sensors include: (measurement,), sliding sensors, vision sensors, infrared sensors, ultrasonic sensors, tactile sensors, etc. Due to the detection objects of each sensor, it is necessary to determine the consistency of different sensor data. By supplementing the information of different sensors, complete external information can be obtained. Therefore, multi-sensor information fusion can improve the intelligence level of electrical appliances. (2) Information fusion Information fusion is the automatic analysis and synthesis of several information sources obtained according to the action sequence of home appliances using computer technology, according to certain criteria, to complete the information processing task. It is manifested in several aspects: ① Divide information levels and form information databases. Information fusion completes the selection process of multi-source information at several levels. Each level represents different levels of information. The essence of information fusion is a combination of multiple information from low level to top level. (3) Information fusion between multiple sensors (through microcontroller). The hierarchical and modular automatic target recognition framework proposed in this paper is decomposed into relatively independent modules according to function, which can provide a variety of different algorithms for system implementation. Multi-sensor information fusion technology promotes the transformation towards intelligence and autonomy by coordinating the use of multiple sensors. It integrates the locally incomplete data provided by multiple homogeneous or heterogeneous sensors distributed in different locations with relevant information from associated databases, eliminating potential redundancy and contradictions between multiple sensors, complementing each other, determining the uniqueness of the current action, and obtaining a consistent description of the object or environment. It merges information from multiple sources through certain algorithms to generate more reliable and accurate information, and makes reliable decisions based on this information. Its key technologies can perform calibration, correlation estimation, error pattern recognition, and state decision processing on various types and individual raw information. Multi-sensor systems are the hardware foundation of information fusion technology, multi-source information is the processing object of information fusion, and fusion algorithms are the core of information fusion. A general method of multi-sensor information fusion is shown in Figure 1. Due to the complexity and diversity of its applications, multi-sensor information fusion currently employs methods such as fuzzy logic, neural networks, and wavelet transform, with neural networks being one of the most important methods. Multi-sensor information fusion technology plays a crucial role in promoting the intelligent and autonomous development of robots. It involves coordinating the use of multiple sensors to integrate relevant information distributed across various related databases. This eliminates redundancy and contradictions that may exist between the local incompleteness provided by multiple homogeneous or heterogeneous sensors located in different locations, reducing uncertainty and making it one of the key technologies for robot intelligence. Its key technologies include calibration, correlation estimation, error pattern recognition, and state decision processing for various types and types of raw information. It merges information from multiple sources through certain algorithms and makes reliable decisions based on this information. Multi-sensor systems are the hardware foundation of information fusion technology, multi-source information is the processing object of information fusion, and the fusion algorithm is the core of information fusion. 3. Intelligent Systems Based on Neural Network Information Fusion Artificial neural networks are network systems with parallel computing capabilities, consisting of many units, also known as neurons, interconnected according to a certain topology. They possess strong nonlinear fitting capabilities and the ability to process multiple inputs and multiple outputs simultaneously. ① Information can be distributed and stored, with large capacity and good fault tolerance; ② Self-learning, self-organizing, self-correcting, and adaptive; ③ The behavior of a neural network is the collective behavior of a large number of neurons; ④ Neurons can handle some very complex environments, unclear knowledge backgrounds, and ambiguous reasoning rules. A neural network is a highly nonlinear dynamic system. Artificial neural networks have the following characteristics: ① Parallel distributed information processing (PDP); ② Learnability; ③ Robustness and fault tolerance; ④ Generalization ability. The greatest advantage of using artificial neural networks for information fusion lies in: large-scale parallel processing and distributed information storage, good adaptability and self-organization, and strong learning, associative, and fault-tolerant capabilities. 4. Neural Network Model A neural network has input/output layer nodes and hidden layer nodes. After passing through the action function, the output signal of the hidden layer node is transmitted to the output layer node, processed, and then output. In execution, the input represents the existence and degree of the target being executed, and the output is the learning pattern. We employ a BP neural network, consisting of three feedforward layers: an input layer, an output layer, and a hidden layer. The input signal propagates forward to the hidden layer, is processed by the nonlinear functions of the nodes, and then reaches the output layer to obtain the corresponding output. If there are discrepancies in the network outputs, the overall error is minimized by adjusting the connection weights. 4.1 Learning Method This method reflects the gradient direction of E in the weight space {W}, and its calculation process is the backpropagation process of the output error through network connections. The feedforward neural network learns through error correction. The learning process consists of forward propagation and backward propagation. During forward propagation, the input signal is processed layer by layer from the input layer through the hidden layers and then propagated to the output layer. The state of each neuron only affects the state of the next neuron. If the output layer cannot obtain the expected output, backward propagation is initiated, returning the error signal of the output signal along the original connection path. This process repeatedly adjusts the connection weights between neurons in each layer of the network to minimize the error signal, reflecting the regularity hidden in the samples in the connections between the network neurons. Once the learning is complete, the neural network possesses the ability to judge sample combinations. The neural network module obtains necessary learning samples and diagnostic results from the expert system's reasoning mechanism and fuzzy processing, and stores the learning results in the neural database. 4.2 Application in home appliance selection: The learning process consists of four steps: First, randomly selecting sensor values ​​for training. Second, learning and training using algorithms to build a neural network. Third, for newly added faults, adjusting the weights on the existing neural network to perform new learning. Fourth, validating the built neural network with other fault values; if the accuracy does not meet the requirements, adding a certain proportion of signals that cannot be correctly judged to the highest priority, and returning to step three for learning until the judgment accuracy meets the requirements. This paper selects a subset of 6 for system learning and testing. The neural network generated after algorithm learning, after training on the test set, meets the requirements . Conclusion: Currently, it is only suitable for selecting refrigerators, air conditioners, rice cookers, microwave ovens, induction cookers, and home cameras. Multi-sensor information fusion technology is one of the key technologies. With the improvement of sensor technology and microcontroller fusion technology, the ability to acquire environmental information and the system's decision-making ability will be continuously improved. Based on the fusion and selection of information from multiple sensors using neural networks, the intelligentization and autonomy of home appliances are promoted, achieving the goal of complete control. References: [1] Wang Yaonan. Intelligent Control System [2] Chen Zhiqiang, Yan Zhilin. Fuzzy Expert System for Quality Diagnosis [3] Wen Chuanyuan. Artificial Neural Network and Its Application [4] [1] Zou Lihe. Digital Signal Processing [5] Peng Xiaojun, Liu Guangbin. New Developments in Circuit Diagnosis Technology [6] He Jiazhou, Zhou Zhihua et al. A Neural Network Method for System Fault Diagnosis Author Introduction : Jin Congying (1956-), female, senior experimentalist, mainly engaged in artificial intelligence research.