Abstract: This paper proposes a distributed motion control system suitable for humanoid walking robots. The system utilizes a CAN bus as the communication tool, a DSP chip as the main processor, and a dual-position sensor feedback structure, combined with a two-layer feedback control method involving a main control computer and a lower-level controller. The entire control system is flexible, powerful, stable, and reliable, significantly improving the motion performance of humanoid robots. Keywords: Humanoid robot; Motion controller; CAN bus; DSP I. Introduction Robotics research is one of the most complex and challenging topics in the field of automation. It integrates multiple disciplines such as mechanics, electronics, computer science, materials science, sensors, and control technology, representing a concentrated embodiment of high-tech achievements across multiple disciplines. Research on humanoid walking robot technology is at the forefront of robotics research, representing, to a certain extent, a country's high-tech development level. Motion control systems are the core of robot control technology and one of the key technologies in robotics research, holding a crucial position in robot control. Therefore, research institutions prioritize the research of robot motion control systems. Coordinated movements, a certain level of intelligence, and the ability to achieve wireless real-time walking have become the themes of current robot development. With the continuous development of modern high technology, represented by electronic computers and digital electronics, especially the widespread application of high-speed digital signal processors (DSPs) and large-scale programmable logic devices (CPLDs and FPGAs), robot motion control systems have evolved from simple structures and functions to structured, standardized, modular, and highly integrated systems. Adopting an open architecture has become an inevitable trend in this technological development. This paper, in line with this trend, designs a multifunctional distributed humanoid robot motion control system. II. Control Object and Requirements We take the latest generation of humanoid walking robot (its appearance is shown in Figure 1) developed by the Robotics Teaching and Research Laboratory of the College of Mechanical and Electrical Engineering and Automation at the National University of Defense Technology as our research object. This robot is approximately 1.55m high, weighs approximately 65kg, is battery-powered, requires no external power supply or control signal lines, can walk without cables, and can perform some basic human leg, hand, and head movements, already possessing preliminary human-like physical characteristics. This new humanoid robot has a total of 36 degrees of freedom (as shown in Figure 2), including 12 for the upper limbs, 12 for the lower limbs, 2 for the head, and 10 for the hands; each joint of the lower limbs has a position sensor, and the feet have multi-dimensional force/torque sensors; it has a visual sensing, voice control system, and wireless remote control module; the entire control system and power supply are integrated on the robot body. In order to truly have the characteristics of "humanoid", the control system must be able to complete multiple functions including motion control and planning, visual perception processing, voice recognition, and other environmental perception. Among them, motion control is the key to the entire control system, and it must be able to meet the following requirements: (1) High system integration, small size, light weight, high power, high efficiency, and airborne. (2) Simple connection between various modules, easy to install and maintain. (3) The controller should have good dynamic response and following characteristics, and small steady-state error and static error. (4) The system is integrated on the robot body, and electromagnetic interference is strong, so it must have strong anti-interference ability. (5) Data exchange between various parts must be real-time, effective, accurate and reliable. III. Motion Control System Design Based on the above requirements, we designed a novel control structure based on the CAN fieldbus. The entire control system adopts a centralized management and distributed control approach, divided into three layers according to the structure and function of the control system: the organization layer, the coordination layer, and the execution layer. The organization layer consists of a workstation outside the robot body, mainly responsible for human-machine interaction, wireless communication, voice, vision, and macro instruction generation, falling under the category of intelligent control, which will not be discussed in detail here. The coordination layer and the execution layer are both integrated on the robot body, completing specific control tasks, falling under the category of physical control, which is the control system in our conventional sense. Its specific structure is shown in Figure 3. 1. Main Control Computer Module: The main control computer is required to be small in size and fast in processing speed, typically using a small-board industrial control computer, equipped with an LCD display and a self-made dedicated function keyboard. It mainly performs online motion planning, motion-level motion control, voice interaction control, visual guidance control, and human-machine interaction functions. It receives information from local sensors, generates task planning data for the joint axis system in real time according to certain control algorithms and task requirements, and sends it to each underlying motion controller via the data transmission bus. 2. Communication Module: The main control computer and each controller communicate via a CAN bus. CAN (Controller Area Network) bus is the most widely used fieldbus and the only fieldbus with an international standard. Compared with general communication buses, its data communication has outstanding reliability, real-time performance and flexibility. Its main features are: (1) CAN bus is multi-master mode, and any node on the network can send data to other nodes at any time. (2) Nodes on the CAN bus can be divided into different priorities by identifiers to meet different real-time requirements. (3) CAN bus adopts non-destructive bus arbitration technology, and low-priority nodes do not affect the transmission of high-priority nodes. (4) The communication rate of CAN bus nodes can reach up to 1MBPS within 40m. (5) The number of nodes on the CAN bus can reach 110 in the standard frame format and is almost unlimited in the extended frame format. (6) The message adopts short frame format, with short transmission time and extremely low error rate. (7) The CAN bus communication medium can be twisted pair cable, which has a flexible structure and is easy to connect. The above features of CAN bus make it very suitable for robot control. Therefore, this paper selects CAN bus as the communication tool for robot control system. The specific connection method is as follows: the main control computer is connected to the bus through the CAN bus interface card, and each motion controller is also connected to the bus through the bus transceiver. The number can be increased or decreased according to the actual situation. Since the CAN bus only uses two wires for communication, it greatly reduces the complexity of the system connection and enhances the reliability of the system. (3) Execution layer module The execution layer is located at the bottom layer of the entire control system and is composed of different types of controllers. It is mainly used to control the specific execution process of each motion joint axis. Since the motors of each motion joint are different in model and the weight they bear are different, the requirements for control accuracy are also different. We designed different motion controllers for them. ① Open-loop DSP motion controller The head and upper limb load weight is relatively light, so an open-loop DSP motion controller is used to control each joint of the head and upper limb. These controllers do not need sampling and feedback. They directly receive the control commands sent by the main control computer and then generate corresponding execution commands to send to each joint axis to make it rotate to the corresponding angle. ② Open-loop MCU motion controller The volume and mass of each joint of the hand are very small, so an open-loop MCU motion controller is used for control. These controllers use MCS-51 microcontrollers as processors and can be directly embedded in the palm of the hand. They receive control commands from the main control computer and use their IO pins to generate the required multi-channel pulse control signals to control the movement of each joint of the hand. ③ Closed-loop DSP motion controller All the axes of the legs are composed of DC geared drive motors with zero-position detection, encoder and potentiometer feedback, and multi-dimensional force/torque sensors. The structure is complex, the control is difficult, and the accuracy requirements are high. Therefore, a closed-loop DSP motion controller is used. This part is the key to the entire control system and is also the focus of our research. (4) Control system flow The specific flow of the entire control system is as follows: The system starts running and completes the initialization work; the main control computer sends control commands to the bottom controllers according to the planning and calculation. After receiving the commands, the bottom controllers combine the information fed back by each sensor and generate corresponding execution commands through a certain control algorithm and send them to each joint execution axis. At the same time, the operation status of the bottom axis is uploaded to the main control computer. The main control computer generates new commands according to the new situation and sends them to each controller. This process is repeated. This is actually two closed-loop feedback processes. The lower-level controller performs small-loop feedback through sensors and each joint axis, while the main control computer performs large-loop feedback through each controller and each joint axis. This allows the robot to have more "intelligence" and better perform offline real-time control. The main control computer sends 200 sets of data to the lower-level controller every second, and the lower-level controller feeds back the same number of data to the main control computer. The maximum communication rate of the CAN bus can reach several thousand frames per second, which is more than enough to meet the control requirements. IV. Detailed Controller Design The closed-loop DSP controller controlling the lower limbs is the core of the entire control system. It bears the entire load weight of the robot, has high output power, and requires high control precision. Therefore, its performance directly affects the realization of robot movement. We specifically designed a closed-loop DSP controller based on dual position sensors, the structure of which is shown in Figure 4. The DSP main processor used is the TI TMS320LF2407A chip, a high-end product in the TI C2000 series, which is very suitable for industrial control. Its two event managers are particularly powerful, designed entirely for motor control, and can directly generate the required PWM pulse control signals using multiple PWM pulse channels. Its CAN bus module can communicate directly with the main control computer without the need for an additional CAN bus controller. An external watchdog can monitor the controller voltage. External memory stores the necessary parameters for the control algorithm. The controller's dual position sensors consist of a voltage output sensor and a photoelectric encoder sensor. The voltage sensor converts the shaft position information into a voltage signal, which is amplified by an amplifier circuit and then converted into a digital signal by a dedicated A/D converter before being sent to the DSP main processor. Instead of using the TMS320LF2407A's built-in A/D converter, a dedicated A/D conversion chip is used to improve conversion accuracy. The TMS320LF2407A's A/D converter can only accept a maximum conversion voltage of 3.3V, while the voltage after power amplification far exceeds this range. Therefore, a dedicated A/D conversion chip is used. Although this circuitry increases the complexity of the controller, it significantly improves conversion accuracy, making it well worth the investment. The encoder sensor converts the axis position information into pulse signals, which are then isolated by an opto-isolator and sent to a dedicated pulse counter. The counted information is then sent to the DSP main processor. The pulse counter uses a popular CPLD device, whose powerful functions greatly improve the controller's performance and also serve as a decoding circuit to provide decoding functionality to the main processor. The main processor analyzes and calculates the received sensor signals to generate corresponding PWM pulse control signals, which are then opto-isolated and amplified before being sent to the underlying axis system to control its operation. Using dual sensors significantly improves feedback accuracy; both signals can be considered simultaneously, or one can be primary while the other provides supplementary and reference signals. The main processor communicates with the main control computer via a CAN bus, receiving commands and feeding back underlying information to achieve higher-level feedback control. The main processor connects to the CAN bus via a transceiver; opto-isolation is required in between to improve accuracy. This controller is directly installed inside the humanoid robot's body. Each controller can simultaneously control six joint axes; only two controllers are needed to control the entire lower limb's motion. V. Conclusion Based on the full absorption of the high-tech achievements of related disciplines today, we have designed a humanoid robot motion control system that is fast, stable, highly integrated, flexible in structure, and easy to use. The entire motion control system can be directly embedded into the robot body so as to complete the specified control tasks in actual operation. At the same time, the control system also has strong expansion function and can be easily transplanted to other similar control mechanisms. It is a multi-functional general-purpose control system with broad application prospects. References: [1] Kazuo Hirai. The Honda Humanoid Robot: Development and Future Perspective[J]. Industrial Robot: An International Journal, 1999, 26 (4) [2] Cen Young and Draig G. Pickin. Accuracy assessment of the modern industrial robot[J]. Industrial Robot, 2000, 27 (6) [3] Tarun Kanti Podder, Nilanjan Sarkar. Fault-tolerant control of an autonomous underwater vehicle under thruster redundancy[J]. Robotics and Autonomous Systems, 2001, 34 [4] Fang Haijun et al. Autonomous humanoid robot control system based on CAN bus [J]. Robot, 2002, 24 (1) [5] Cao Yu. Design and application of airborne motion controller for humanoid walking robot [C]. Changsha: Archives of the School of Mechanical and Electrical Engineering and Automation, National University of Defense Technology (internal material), 2002 [6] Shi Zongying et al. Humanoid robot control system based on fieldbus [J]. Engineering and Application, 2002, (2) [7] Wu Kuanming. CAN bus principle and application system design [M]. Beijing: Beijing University of Aeronautics and Astronautics Press, 1996 [8] Rao Yuntao et al. Fieldbus CAN principle and application technology [M]. Beijing: Beijing University of Aeronautics and Astronautics Press, 2003 [9] Jiang Simin et al. TMS320LF240x DSP hardware development tutorial [M]. Beijing: Machinery Industry Press, 2003 [1] Huang Zhengjin et al. Introduction and Application of CPLD System Design Technology [M]. Beijing: Electronic Industry Press, 2002.