Abstract: In recent years, the research and design of indoor mobile robots has become a focus of attention. We adopted a microcontroller as the core controller of the robot and developed a robot experimental platform by jointly using ultrasonic sensors, collision sensors, stepper motors and their control chip Ta8435. Finally, we introduced fuzzy control and fuzzy neural networks, and used fuzzy control and fuzzy neural network technologies to conduct MATLAB simulation research on fuzzy control obstacle avoidance and fuzzy neural network path tracking in indoor robot navigation, achieving the expected goals. Keywords: Indoor robot, fuzzy control, fuzzy neural network, trajectory tracking 1 Introduction With the technological advancements in sensors, electronics, mechanics and materials, robots have opened up new fields in the service industry beyond manufacturing for the first time. Indoor robots are a new type of service robot. This robot can move automatically in a room and perform tasks such as vacuuming and home security. It integrates many disciplines such as mechanics, electronics, sensor technology, computer technology, control technology, robotics, and artificial intelligence. The development of the home service robot industry has also promoted the development of mobile robot technology, image and speech recognition, sensors and other related technologies. 2 Overall Design of the Robot 2.1 Robot Design Requirements The overall functional requirements of the robot we designed are as follows: After the power button is turned on, the robot walks in a straight line and then covers the ground according to the stored electronic map. Real-time obstacle avoidance is performed during the movement. The robot stops working after the coverage area is completed. Based on the above functions, the robot should also meet the following basic requirements: (1) The robot can accurately detect obstacles around the robot, and the detection height should be limited to avoid interference from the ground; (2) The robot's movement speed needs to be controlled more precisely. The mobile robot needs to complete obstacle avoidance and path planning tasks on the basis of autonomous movement. Therefore, the robot must have a clear understanding of its surrounding environment. The mobile robot system we designed mainly uses the Cygnalf 020 microcontroller as the controller, selects the stepper motor and its drive control circuit as the power source, uses ultrasonic sensors as the means of detecting the external environment, and adds collision sensors to compensate for the blind spots of ultrasonic sensors. The C8051F020 microcontroller is a highly integrated system-on-a-chip (SoC). It integrates two multi-channel ADC subsystems (each including a programmable gain amplifier and an analog multiplexer), two voltage output DACs, two voltage comparators and a voltage reference, an SMBus I2C bus interface, a UART, an SPI bus interface, five general-purpose 16-bit timers, a programmable counter/timer array with five capture/compare modules, an internal oscillator, eight 8-bit general-purpose I/O interfaces, 64KB of FLASH program memory, and an 8051-compatible high-speed microcontroller core. 2.2 Robot Hardware Structure Diagram The structure of our designed indoor robot is shown in Figure 1: it is a three-wheeled vehicle with support wheels. The front wheel is a support wheel, serving to support the vehicle body without providing guidance. The two rear wheels are drive wheels, driven by stepper motors. The vehicle's speed and rotation angle can be controlled by adjusting the speed of the two rear wheels. Each motor and rear wheel forms a speed closed loop, providing constant speed output. Within the working load, the speed of the two rear wheels can be adjusted by regulating the input voltage of the two motors. The chassis uses thin steel plates and a three-wheel support system. The front wheel is a support wheel, and the rear wheels are two parallel drive wheels, driven and controlled by stepper motors and their microstepping control chip TA8435. An ultrasonic sensor and a collision sensor are installed at the front. A Cygnalf 020 microcontroller is placed in the middle for downloading control programs. 2.3 Robot Cart Motion Trajectory The motion path of the robot car is shown in Figure 2. The starting point O of the car's motion is the origin of the coordinate system. After time t, the car moves to point A. Here, Ex(t) and Ey(t) are the displacements of the car in the X and Y directions, respectively; θ(t) and ω(t) are the angular displacement and angular velocity of the car; and υ is the moving speed at the midpoint between the two rear wheels. Figure 1. Car structure diagram. Figure 2. Motion path diagram. 2.4 Ultrasonic ranging circuit design. Figure 3. Ultrasonic receiving circuit diagram. The ultrasonic circuit structure is shown in Figures 3 and 4. We use an ultrasonic sensor with a center frequency of 40kHz, thus determining that the center frequency of the ultrasonic sensor excitation signal is also 40kHz. Since the ultrasonic sensor is highly sensitive to changes in signal frequency, a slight deviation from the rated center frequency will cause a significant attenuation of the sensor's emitted sound intensity. Therefore, generating the required rated excitation signal and ensuring the relative stability of the excitation signal's center frequency is the main task of the ultrasonic oscillation signal generation circuit. 2.5 Collision sensor. During the actual debugging process, due to the measurement blind zone between ultrasonic sensors, a collision sensor was designed at the front end of the robot. A microswitch is installed on each of the left and right sides of the collision plate. When no collision occurs, the sensor sends a high-level signal to the microcontroller. When the moving robot collides with the front, it will trigger the left and right switches to respond and send a low-level signal to the controller, thereby controlling the robot to make a corresponding reaction action. Figure 4. Ultrasonic transmitting circuit diagram. Figure 5. Microswitch connection principle diagram. 2.6 Stepper motor control based on TA8435 chip. The two pins M1 and M2 determine the motor's rotation mode. M1=0, M2=0, the motor operates in full-step mode; M1=1, M2=0, the motor operates in half-step mode; M1=0, M2=1, the motor operates in 1/4 microstep mode; M1=1, M2=1, the motor operates in 1/8 microstep mode. CW/CWW controls the motor's rotation direction. CK1 and CK2 are clock inputs, with a maximum frequency not exceeding 5kHz. Controlling the clock frequency controls the motor's rotation speed. When REFIN is high, NFA and NFb output voltage is 0.8V; when REFIN is low, NFA and NFB output voltage is 0.5V. These two pins control the stepper motor's input current. With O and C lines unconnected, the stepper motor operates as a 2-phase bipolar motor, which improves the stepper motor's output torque. 3. Design of Fuzzy Neural Network for Mobile Robot Trajectory Tracking In the research of mobile robot trajectory tracking, neural network methods or fuzzy neural network methods described in other literature are commonly used. Fuzzy control systems cannot adapt to changes in the control process, affecting the control effect, while neural networks have the ability to adjust online. Based on the advantages of fuzzy logic technology and neural network technology, fuzzy technology and neural networks are organically combined to form a fuzzy neural network control system, which can realize automatic extraction of fuzzy rules, automatic generation of fuzzy membership functions, and online adjustment of parameters. Therefore, we adopted the Gaussian fuzzy neural network method to design a fuzzy neural network controller, which does not depend on the precise mathematical model of the system and can complete the required robot trajectory tracking task. 3.1 Network Structure The outputs of a multiple-input multiple-output (MIMO) system are independent of each other, so the MIMO fuzzy rules can be decomposed into multiple multiple-input single-output (MISO) fuzzy rules. Therefore, without loss of generality, only the MISO fuzzy neural network is introduced below. The structure of the fuzzy neural network in this paper is shown in Figure 6: Figure 6 Fuzzy Neural Network Model In this model, there are four layers, namely the input layer, the membership function generation layer, the inference layer, and the defuzzification layer. The input layer consists of P nodes. In the membership function layer, each input node is divided into m word sets, where m is obtained by clustering the input samples using the K-means method. The value of m can be adjusted according to actual needs. 3.2 Description of Input and Output of Each Layer of the Fuzzy Neural Network The improved fuzzy neural network is a four-layer network. The layers are: the first layer introduces the input into the network; the second layer fuzzifies the input variables using a Gaussian membership function; the third layer corresponds to fuzzy inference, using a product to represent the fuzzy AND operation; and the fourth layer corresponds to the defuzzification operation. The number of nodes m in the inference layer is obtained by clustering the samples using the K-means method, and the value of this parameter can be adjusted according to actual needs. 4 Simulation Experiment and Results The simulation program in this section is implemented in C language and run in the MATLAB simulation environment. We use three input variables, i.e., p = 3, representing the distance data collected by the three-direction (front, left, right) ultrasonic ranging system used by the mobile robot. Each input variable is divided into five word sets, i.e., m = 5. The training samples consist of 40 discrete data sets, i.e., N = 40. The training iterations were 3000. The learning rate was set to 0.0002. Initial center and initial width were both random numbers between [1, 2], and initial weights were set to {80, 50, 50, 35, 35}. The expected output, actual output, and their error are shown in Figure 7. The membership function curves of the three input variables before and after training are compared in Figures 8, 9, and 10, respectively. As can be seen from the figures, the fuzzy neural network provides a good tracking effect, with a mean square error of 0.132 after training. Figure 7 Expected output, actual output, and their error Figure 8 Figure 9 Figure 10 Figure 8, 9, 10 Comparison of membership function curves of the three input variables before and after training 5 Conclusion The design of the indoor service robot involves the comprehensive application of many disciplines, including microcontroller technology, ultrasonic sensor ranging technology, GPS positioning technology, collision sensor technology, stepper motor control technology, neural networks, fuzzy control, and fuzzy neural networks. Leveraging the online learning capability of fuzzy neural networks, a fuzzy neural network controller was designed to track the predetermined trajectory curve of a mobile robot. MATLAB simulations verified that it meets the requirements for robot trajectory tracking. The authors' innovations include: using a microcontroller as the core controller of the robot, and jointly developing a robot experimental platform using ultrasonic sensors, collision sensors, stepper motors, and their control chip Ta8435. Furthermore, MATLAB simulations were conducted on fuzzy control obstacle avoidance and fuzzy neural network path tracking in indoor robot navigation using fuzzy control and fuzzy neural network technologies, achieving the expected objectives. References: [1] Hu Yueming, Ding Weizhong, Wu Xinsheng. Research status and prospect of vacuuming robots[J]. Robotics Technology and Application, 2002(l):33-37 [2] Li Kaisheng, Zhang Huihui, Fei Renyuan, Zong Guanghua. Development dynamics and prospects of foreign service robots[J], Manufacturing Automation, 2000, 22(6):1-4 [3] Luo Bencheng, Yuan Kui et al. Development of multi-channel ultrasonic environmental detector for robots[J]. Journal of Graduate School of Chinese Academy of Sciences, 2002(2):172-176 [4] Liu Chong, Xiao Nanfeng. Design and implementation of intelligent security and home service robots[J]. Microcomputer Information, 2006, 2-2: P211-214, P74 [5] Chen Wentao, Zhu Zhijian. Research on microcomputer-controlled stepper motors[J]. Electronics and Packaging, 1997, 5(8):36-37 [6] Wu Haibin, Zhu Shiqiang, Ma Xiang. Obstacle avoidance and path planning of autonomous vacuum cleaner robot in unstructured environment [J]. Robot, 2000, 22(7) [7] Zhang Liangjie, Li Yanda. Research on automatic design system of intelligent controller based on fuzzy logic and neural network technology. Technical report of Department of Automation, Tsinghua University, 1994 [8] Zhang Haoran, Han Zhengzhi, Li Changgang. Neural network identification and control toolbox based on MATLAB [J]. Computer Simulation, 2003(3)