1. Introduction Wearable lower limb assistive robots are a type of assistive rehabilitation robot, a device designed to help people extend their lower limb motor abilities. Their basic principle is based on human motor behavior awareness information, with servo motors installed at the leg joints (hip and knee joints) driving joint movement. By changing the angles and speeds of each joint, they achieve coordinated movement with the human leg and provide assistance, reducing the intensity of movement under weight-bearing or prolonged walking. They offer treatment and correction for individuals with abnormal motor behaviors, forming a coordinated and perfect whole with the human body. Currently, approximately ten laboratories worldwide are engaged in research on wearable human assistive robots, with Japan and the United States leading the way. There are no related reports in China. In 2002, Tsukuba University in Japan developed a robotic hybrid assistive limb (HAL). The mechanical exoskeleton is strapped to both sides of the leg, using EMG sensors attached to the skin of the leg to detect muscle electrical currents and control electric motors to drive the mechanical exoskeleton to assist leg movements. The Robotics and Human Engineering Laboratory at the University of California, Berkeley, has developed the U.S. military's "Berkeley Lower Extremity Exoskeleton" (BLE-EX), consisting of a backpack-style exoskeleton, metal legs, and corresponding hydraulic drive systems. The mechanical system employs a design similar to a humanoid structure. The backpack-style exoskeleton allows the operator to carry a certain load, with the effective force transmitted directly to the ground via the exoskeleton without passing through the wearer. The lower limb exoskeleton can carry external loads and its own weight (including the operator's weight) for long distances on rugged terrain, enhancing the load-bearing capacity and increasing the marching speed of fully armed soldiers. However, these devices share common drawbacks. Because electromyography (EMG) sensors infer a person's behavior and consciousness based on weak electrical signals transmitted through the skin surface during muscle activity or the firmness of muscles, most of the sensors must be in direct contact with and adhered to the skin, requiring special fixing devices. This directly leads to inconvenience in wearing them. Furthermore, human sweat secretions and the quality of sensor installation affect the stability and accuracy of the acquired information. Moreover, the large and complex amount of information is susceptible to interference, increasing the difficulty of control. Therefore, this paper designs a novel wearable lower limb assistive robot sensing system. This system is used to acquire the contact force between the human lower limb and the robot exoskeleton, and uses this force information and joint angle information to control the robot exoskeleton to assist the human lower limb movement. 2. Assistive Robot System The wearable lower limb assistive robot mainly consists of three parts: mechanics, sensing, and control. The robot exoskeleton contains 12 degrees of freedom, with 6 degrees of freedom for each leg, 3 degrees of freedom for the hip joint, and 1 degree of freedom each for the knee, ankle, and foot joints. This design not only meets the design requirements of previous anthropomorphic robot walking mechanisms but also achieves the design requirement of coordinating with the human leg movement without causing motion interference, as shown in Figure 1. The execution part mainly refers to DC servo motors; the system requires four, which are fixed to the hip and knee joints of both legs respectively. The wearable lower limb assistive robot control system mainly uses a PC104 embedded control system board and a PC104 CAN card. The control structure of the entire system is shown in Figure 2. 3. Robot Perception Module 3.1 Human-Machine System Contact Information Wearable lower limb assistive robots mainly utilize human lower limb motion information to provide assistance. This motion information mainly includes leg contact force signals, foot force signals, and knee and ankle angle signals between the human body and the exoskeleton robot. To acquire this motion information, a multi-sensor perception system based on the CAN bus was designed, solving the problems of single master node and poor real-time performance in traditional sensor communication methods (mainly RS-232 and RS-485). The system consists of a motor encoder, two two-dimensional force sensors mounted on the legs, and six one-dimensional force sensors mounted on the soles of the feet. The leg force sensors are fixed to the upper part of the knee and ankle joints to measure the contact force between the human body and the exoskeleton; the foot force sensors are mounted on the toes and heels to measure the ground reaction force; the motor encoder is used to measure the rotation angles of the hip and knee joints, as shown in Figure 1. 3.2 System Design The two-dimensional leg force sensors are used to measure the magnitude of the contact force between the robot exoskeleton and the human body. The accuracy and stability of their measurement are of great significance to the control of the assistive robot. The leg force sensor mainly consists of two two-dimensional force sensors used to measure the contact force between the human thigh and calf and the robot's exoskeleton. This contact force includes the force along the human leg (X direction) and the force perpendicular to the leg (Y direction). In the control of the lower limb assistive robot, in addition to knowing the force between the human leg and the robot, it is also necessary to know the force exerted by the human foot on the robot. The foot force sensor measures the reaction force of the ground on the human-machine system. The point of force exerted by the human foot on the ground can be represented by three support points, located at the root of the first metatarsal bone, the root of the fifth metatarsal bone, and the heel. The human body is supported by the arch of the foot generated between these three points, and the body weight is transmitted to the ground through these three points. In order to accurately obtain the force information of the foot during walking, the foot force sensors are installed at these three points. Three one-dimensional force sensors are required for each foot, for a total of six one-dimensional force sensors. The specific installation positions are shown in Figure 3. Due to limitations in the foot mechanism, the sensor elastomer is relatively small, with a body size of φ40 mm (diameter) × 8 mm (thickness) and a measuring range of 1000 N. 3.3 Sensor Design The design of the elastomer is crucial in multidimensional sensor design. Based on the simulation analysis of the static and dynamic characteristics of the sensor elastomer using the finite element method, this paper designs an elastomer structure based on an E-type diaphragm. This sensor structure features simple structure, high sensitivity, low interdimensional coupling, and easy calibration. The entire elastomer mainly consists of three parts: an elastic diaphragm, a strain gauge, and a force-bearing adapter. The elastomer is composed of two E-type diaphragms, enabling the measurement of strain forces in both the X and Y directions. The elastic diaphragm is circular, with a diameter of φ15 mm and a thickness of 2 mm, with the thickness direction aligned with the measurement direction. The sensing element is a foil resistance strain gauge, which is attached to the E-type diaphragm. The sensor output is the stress on the E-type diaphragm. While there are many methods for measuring stress, this study uses a foil resistance strain gauge attached to the E-type diaphragm to measure the magnitude of stress on the elastomer. The strain gauge patch positions are shown in Figure 4. Strain gauges in the X and Y directions are installed at the lower end of the E-type diaphragm. The four strain gauge resistors form a Wheatstone bridge circuit (as shown in Figure 5) to achieve automatic decoupling of the output signal. When a force is applied to the sensor, the stress on the sensitive resistor varies in different directions due to differences in the magnitude and direction of the force, thus obtaining the relationship between force and strain. Taking the X direction as an example, the annular flat diaphragm of the sensitive elastic part of the E-type diaphragm is a thin plate structure. Under the action of the X-direction force, the boundary conditions are relatively simple and can be equivalent to a circular thin plate with a fixed outer circumference and concentrated stress acting on a hard center. According to thin plate theory, the radial stress εr and tangential stress εθ of the diaphragm with a fixed periphery and a hard center at radius r are given by the following formula: where ω and h are the normal displacement and thickness of the circular diaphragm, respectively; F is the equivalent concentrated force of the applied force; f(r) and P(r) are functions only related to r. From the above formula, it can be seen that when the radius r is constant, that is, when the position of the piezoresistor is fixed, the strain ε on the surface of the circular diaphragm is ε=kFF (3) where kF is the strain coefficient constant. Since an equal-arm bridge is used, that is, R1=R2=R3=R4, we have the following formula: ε1, ε2, ε3, ε4 are the strains of the four sensitive resistors R1, R2, R3, R4 respectively; ε is the total strain of the circular diaphragm; G and k are constants; UX is the bridge output voltage. Combining formulas (4) and (5), the output voltage signal of the bridge is proportional to the force signal of the sensor. Measuring the output voltage signal can obtain the force signal of the target. The sensor hardware circuit adopts an embedded on-chip system, which consists of two parts: digital circuit and analog circuit. The analog circuit consists of a signal zeroing circuit, an operational amplifier circuit and an analog filter circuit. The digital circuit mainly includes an A/D sampling module, a digital calculation module, a CAN bus controller, a CAN bus driver and necessary peripheral circuit modules. Figure 6 is the hardware circuit principle block diagram of the data acquisition and processing system of the assistive robot force transmission system. The software design is divided into lower-level (microprocessor) software design and upper-level (PC) software design. Each sensor is interconnected as a node via a CAN bus. Upon receiving a command from the upper-level computer, the sensor first performs command judgment and then performs corresponding data processing based on the different commands. The upper-level (PC) mainly includes commands such as zeroing, force information (digital quantity), force information feedback, force information query, and alarm masking. The lower-level software design mainly consists of three parts: a data acquisition program (A/D conversion), a data processing program, and a CAN bus communication program. Before initiating the CAN interrupt, a data acquisition is performed in the main program to obtain the initial values ​​of the sensor system, including three A/D conversion channels. A delay is then applied to complete the channel initialization. Data acquisition is completed within the CAN interrupt program; each interrupt acquires one set of three-dimensional force information data and performs the corresponding A/D conversion. Simultaneously, the conversion results are read and digitally processed. Digital processing mainly consists of two parts: digital filtering and force information decoupling. Digital filtering primarily uses a combination of window moving method and data averaging method. After decoupling, the data is sent to the CAN bus via the SendData() function. The host computer receives the data from the slave computer by recognizing the ID number. The specific process is shown in Figure 7. 4. Sensor Calibration Experiment. The complexity of the E-type diaphragm element structure makes it more difficult to guarantee the consistency of product characteristics than with one-dimensional sensors. The patching process of strain gauges is difficult to guarantee an absolutely ideal situation. These factors determine that there is a certain error between the actual static characteristics of the sensor and the theoretical calculation values. Therefore, the static characteristics of the sensor are generally obtained through calibration experiments, and the accuracy of the calibration directly affects the measurement accuracy of the sensor during use. The calibration of a sensor is to establish the quantitative relationship between the three output values ​​of the sensor and the three-dimensional force acting on the origin of the sensor coordinate system. The calibration experiment includes two parts: static calibration and real-time measurement verification. In order to reduce the influence of random errors, a least squares calibration method with a certain redundancy force vector is adopted. Let F be the loading force matrix, V be the output matrix (digital quantity) of the sensor, C be the calibration matrix, and E be the error matrix, then F = CV + E (6) Where: F and V are known quantities; E can be set. Thus, the solution of the calibration matrix can be transformed into solving the calibration matrix C, so that equation (6) is optimal in the sense of least squares. In the calibration process of the miniature finger force sensor, the relationship between the load applied to the sensor in the X and Y directions and the sensitive bridge circuit is measured. The correspondence between the measured value (digital quantity) and the value of the added weight is shown in Figure 8 (XLable represents the load applied to the sensor calibration, and Yable represents the digital quantity output by the sensor). As shown in Figure 8, when force is applied to the sensor in the X direction, the mapping relationship between the load and the output of the sensor's sensitive bridge circuit can be considered basically linear, and the maximum coupling in the Y direction does not exceed 2.5%. Using the least squares method, two sets of static calibration matrices for the sensor are obtained. From this, the Type I error of the sensor can be calculated to be 2%, and the Type II error to be 2.5%. Real-time measurement verification of the sensor using the two sets of calibration matrices C1 and C2 shows that the maximum Type I error does not exceed 2%, and the Type II error does not exceed 2.5%. The static calibration matrices obtained through this calibration system are close to the theoretical design values, indicating that the calibration system and calibration scheme are feasible. 5. Conclusion This paper designs a lower limb motion information sensing system based on a CAN bus for a novel human-assisted rehabilitation robot, as shown in Figure 9. Analysis of the control information required by wearable assistive robots determines the types, quantities, and installation locations of sensors. This paper focuses on the elastomer design, measurement circuits, and upper and lower computer software for the leg and foot force sensors. Calibration experiments were conducted on the sensors, and the data was analyzed to provide general performance indicators. The results demonstrate that the design theory and process in this study are correct and can basically meet the needs of the wearable lower limb assistive robot control system. Future work will focus on the following: ① Further improving the elastomer structure of the sensors, further reducing the volume of the elastomer and accurately determining the strain gauge patch positions while meeting the sensor performance indicators; ② Improving the sensor measurement circuit design, adding filtering circuits, and improving the amplification circuit; ③ Improving the sensor calibration system to minimize calibration errors.