Design and Control of a High-Dynamic Modular Reconfigurable Legged Robotic System

Here, we present a protocol to enable beginners to replicate a highly dynamic bipedal robotic system and a modular quadrupedal robot composed of bipeds as the smallest units.

Legged robots possess exceptional terrain adaptability, making them an ideal platform for outdoor exploration and cargo transport across complex terrains. The number and configuration of legs play a crucial role in their performance; however, most current designs are monolithic, lacking the flexibility for reconfiguration. The protocol presented in this paper details the design and fabrication of a modular legged robot system with robust walking capabilities and flexible reconfiguration between bipedal and quadrupedal configurations. First, various functional modules are mechanically assembled to complete the construction of the bipedal platform and the assembly of the modular connecting device. Subsequently, using debugging software, we configured the inertial measurement unit and motors, including set controller area network (CAN) identity (ID), baud rate, and other working parameters to ensure they are in the correct working state. Then, we designed a whole-body control strategy and a distributed control framework to ensure stable walking of the bipedal unit and coordinated movement of the assembled structure, respectively. Finally, we validated the system's effectiveness on both bipedal and quadrupedal configurations, enabling the robot to achieve stable outdoor walking.

Legged robots, utilizing a discrete support mechanism, exhibit superior terrain adaptability and agility^1,2. In recent years, legged robots have been deployed in scenarios such as rescue and search operations, achieving favorable results^3,4,5. Legged robots of various configurations offer distinct advantages for different tasks. Bipedal robots, with their simple design, can navigate narrow spaces but have limited load capacity. Quadrupedal robots are more complex but can carry heavier loads and move faster. Robots with six or more legs provide greater stability but are harder to maintain. However, existing legged robots generally employ an integrated design characterized by a singular configuration that lacks the flexibility to adapt or switch between different configurations^6,7,8. This design approach results in a high incidence of faults and presents significant maintenance challenges, as single points of failure may lead to system collapse. In contrast, robots designed with the modular assembly concept can easily reconfigure to meet different task requirements^9,10. When facing rugged terrain, they can enhance their traversability by increasing the number of legs or modules¹¹.

Currently, research on modular legged robots is still in the exploratory stage, primarily focusing on small or miniature-legged platforms, which mainly move by crawling or slithering^{11,12,13,14,15}. Some assembly sub-modules even lack independent movement capabilities and can only achieve mobility when combined^16,17. This makes them difficult to deploy in practical scenarios for tasks such as transportation and reconnaissance. To address these challenges, this paper proposes a modular-legged robotic system with flexible mobility capabilities. Distinct from previous methods, the legs of the assembly sub-modules in this paper are inspired by the mammalian leg configuration, providing rapid running capabilities. We select the bipedal robot as the smallest splicing module due to its simple structure, stable walking capability, and more human-like walking style^18,19.

Additionally, some of the modular legged robots mentioned above use latches or fasteners to connect modules¹², weakening their rapid connection and disconnection capabilities. To simplify the connection process and avoid the use of clips and fasteners, we employ an electromagnetic attachment mechanism between modules, with controllable and flexible activation/deactivation of the magnetic force. To fully leverage the advantages of the modular legged robot, we employ a distributed control approach to manage the locomotion of the robot discussed in this paper. The effectiveness of the constructed system and the control method has been validated through prototype experiments. The proposed system may assist in addressing the needs of large-scale material transport in unstructured terrains or rapid reconnaissance in unknown environments.

The objective of this protocol is to thoroughly present the design and manufacturing process of the system described in this paper, enabling interested parties to replicate or create robots with similar functionality to meet their needs. The bipedal module in this paper is based on our previous work. We have installed a docking mechanism on it and performed parameter fine-tuning²⁰.

1. Construction of the machine

1. Assemble the linkage, housing, and calf using screws and fasteners, and then connect the assembled linkage to the motor output shaft's key pin using fasteners (Figure 1).
   NOTE: The leg, which comprises the linkage, motor, bearings, etc., is the primary component of the robot responsible for its motion.
2. Initialize the motors using the software provided by the motor manufacturer. This includes the calibration of Hall sensors and encoders.
3. Fasten the two assembled legs with carbon fiber plates and appropriate connectors to create a bipedal structure, as illustrated (Figure 2).
4. Fix the carbon fiber tube to both sides of the docking device with screws and secure the electromagnet in its central groove (see Figure 3).

2. Building the control box

1. Assemble the power board based on the circuit schematic and physical diagram shown in Figure 4.
2. Position and secure the power management module, inertial measurement unit (IMU), microcontroller unit (MCU), and other components in their designated locations as illustrated (Figure 5).
3. Connect the universal serial bus (USB) port of the MCU to the IMU using the JT30-USB connector.
4. Link the Ethernet port of the MCU to the router using an Ethernet cable.
5. Use wiring to connect the motor switch, MCU switch, and main switch to their corresponding interfaces on the power management module.
6. Connect the serial peripheral interface (SPI) board's CAN1 and CAN2 interfaces to the power management module's CAN1 and CAN2 interfaces, respectively.
7. Link all motors to the power and CAN interfaces on the power management module via their corresponding interfaces.
   NOTE: The left leg's motors are linked to the left-side power and CAN interfaces on the power management module via the left-side connector on the control box; the right side is connected similarly.

3. Debugging the motors

1. Open the motor debugging software. Connect the computer to the debug motor with a USB-CAN module.
2. Assign the motor's CAN ID. Set the value of ab/ad to 1, hip to 2, and knee to 3.
3. Set the joint zero position and positive direction sequentially, as shown in Figure 6.

4. Control scheme

1. Establish the kinematic transmission relationship between the robot's links as given by
   
   NOTE: The z axis aligns with the joint axis i while the axis aligns along the common perpendicular between joint axes i and i + 1, directed from joint i to i + 1. If the axes intersect, x_i is perpendicular to the plane of intersection. a_i-1 is the rotation angle about x_i-1 from z_i-1 to z_i; a_i-1 is the distance along x_i-1 from z_i-1 to z_i; θ_i is the rotation angle about z_i from x_i-1 to x_i; and di is the distance along z_i from x_i-1 to x_i.
2. Establish full body dynamic equation given by
   
   NOTE: J_c and f_c denote the Jacobian matrix mapping foot forces to joint space and the foot contact force, respectively. M(q) is the inertia matrix, represents centrifugal and Coriolis terms, and g(q) is the gravitational force on the robot.
3. Establish whole-body balance control equations.
   NOTE: The control law for the bipedal module integrates virtual model control (VMC) and whole-body control (WBC). For detailed control equations, refer to a previous work²⁰.

5. Write the program

1. Use ST-LINK to import the binary (BIN) file of the communication program between the MCU and the motor into the SPI board.
   NOTE: The SPI board serves as the communication module between the MCU and the motors, with the STM32 as the control core.
2. Transmit the compiled robot locomotion control program from the PC to the MCU via an Ethernet cable.

6. Starting the bipedal robot module

1. Set the robot to the initial posture, as illustrated (Figure 7A).
2. Launch the executable file compiled from the control code on the MCU.
3. Send the stand command to the robot using the remote control (Figure 7B).
   NOTE: The remote control buttons can be configured as needed. It is recommended that the joystick be used to control speed and the buttons switch between different states.
4. Activate the robot's locomotion mode via the remote control (Figure 7C).
5. Use the remote control's joystick to maneuver the robot forward, backward, and steer.

7. Starting the modulary reconfigured quadruped robot

1. Configure the routers of the two bipedal modules to be on the same network segment.
2. Establish the distributed control framework based on Lightweight Communications and Marshalling (LCM)²¹.
3. Install the assembled splicing device components on the head of the biped robot.
4. Position the two bipedal modules in the initial state as specified and activate the magnetic force of the splicing device.
5. Launch the executable programs of the two bipedal modules separately.
6. Transmit the standing command to the robot via the remote control.
7. Activate the robot's locomotion mode via the remote control.
8. Use the remote control's joystick to maneuver the splicing robot forward, backward, and steer.

To validate the effectiveness of the proposed system, we conducted outdoor walking tests on multiple terrains. Initially, a single bipedal module was selected as the test subject, and motion tests were conducted on both artificial grass and a synthetic track. As shown in Figure 8, the robot demonstrated stable locomotion on both terrains. The posture and joint torque data of the robot during locomotion are shown in Figure 9. The amplitude of the robot's roll and pitch angle fluctuations remains within 0.04 rad (the reference is 0), indicating its decent capability in posture control. On the other hand, the joint torque data show no sharp spikes or glitches, with relatively smooth transitions, further confirming the robot's stable operation with only slight vibration.

In the reconfiguration experiment (Figure 11), we assembled a composite quadruped robot using two bipedal units and controlled it to achieve omnidirectional movement through a distributed control approach. Dynamic switching between the bipedal and quadrupedal configurations was successfully achieved by remotely controlling the activation and deactivation of the electromagnetic device's adhesion force. In the snowfield shown in Figure 11B, a single bipedal module occasionally experiences slipping and instability. However, when the two modules are combined, the overall stability is enhanced, allowing for stable locomotion in the snow.

Figure 1: The construction of the leg. (A) Assemble the leg linkages and fasteners. (B) Integrate the motor into the linkage. Please click here to view a larger version of this figure.

Figure 2: The installation of the bipedal structure. (A) The carbon fiber plate and connectors used for fixation. (B) The assembled bipedal lower limb structure. Please click here to view a larger version of this figure.

Figure 3: The installation of the splicing device. (A) The components of the splicing mechanism. (B) The assembled splicing mechanism. Please click here to view a larger version of this figure.

Figure 4: Assembly of the power management module. (A) The schematic diagram of the power management module. (B) The physical image of the module. Please click here to view a larger version of this figure.

Figure 5: Control box internal structure diagram. (A) The internal parts. (B) The schematic of its electrical connections. (C) The physical integration of the control box. Please click here to view a larger version of this figure.

Figure 6: Schematic of the robot's joint zero position. q_i denotes the active degrees of freedom in the robot's legs while L_i represents the lengths of the respective linkages. Please click here to view a larger version of this figure.

Figure 7: Robot startup procedure. (A) Positioned in the initial posture. (B) Control the robot to stand upright. (C) Activate the locomotion mode. Please click here to view a larger version of this figure.

Figure 8: Outdoor locomotion of the bipedal module. (A) Snapshots of the bipedal robot walking stably on artificial turf. (B) Snapshots of the bipedal robot's forward locomotion on a synthetic track. Please click here to view a larger version of this figure.

Figure 9: Data recorded from outdoor bipedal experiments. (A) The walking data on artificial grass. (B) The walking data on a synthetic track. Please click here to view a larger version of this figure.

Figure 10: Block diagram of the distributed control system for the modular coupling robot. Please click here to view a larger version of this figure.

Figure 11: Configuration-switching outdoor locomotion experiment. (A) Quadrupedal configuration grassland locomotion experiment. (B) Quadrupedal configuration snow locomotion experiment. Please click here to view a larger version of this figure.

The modular legged robot proposed in this paper offers a new perspective on legged robot design, where reconfiguration and coordinated movement are achieved through the combination of multiple distributed leg modules. This approach presented in this paper has been experimentally validated, confirming its capability for configuration reconfiguration and coordinated movement. Sections 1, 2, 4, and 5 in this protocol represent the critical stages in the implementation of this design, covering the development of both the robot's mechanical structure and electrical control system.

There is an increasing demand for legged robots with high stability and heavy-load capacity in practical applications. For example, in disaster rescue scenarios, they must carry equipment and supplies over rugged terrains like debris and rubble²². Most legged robots adopt an integrated structural design approach, but this approach faces issues such as poor adaptability to multiple tasks, high failure rates, and low maintenance efficiency. Furthermore, the agile mobility of legged robots and their high load-bearing capacity are difficult to balance simultaneously. In contrast, the modular legged robots offer advantages such as flexible configuration switching and enhanced maintenance convenience due to their modular design concept.

At present, research on modular robots is still in the exploratory stage, though some preliminary results have been achieved. Existing studies on modular legged robots focus on small-scale platforms, primarily utilizing crawling or creeping locomotion with limited mobility. Additionally, most modular robots require the combination of multiple modules to achieve movement, and individual modules have poor mobility and lack autonomous reconfiguration capabilities. The system proposed in this paper offers advantages in terms of movement flexibility and ease of docking. As validated previously²⁰, this bipedal module is capable of executing high-speed running, and its docking method used in this paper does not require the manual installation of pins, fasteners, or other cumbersome operations.

Here, we outline and discuss several typical application scenarios of the method proposed in this paper. For instance, in the industrial and construction sectors, legged robots are required to transport heavy materials or measurement equipment into construction areas, especially in narrow or hazardous zones where traditional machinery cannot access²³. Another typical application scenario is the rapid exploration or reconnaissance of an unknown area. The modular, reconfigurable robot proposed here can leverage its distributed control advantages by decomposing the structure into multiple bipedal modules, enabling collaborative exploration in multiple directions and regions. This can enhance mapping and reconnaissance efficiency.

The key to this protocol is the stable locomotion of the biped module. If it fails to start properly, troubleshooting the following aspects is recommended. First, the motor debugging software provided by the manufacturer is used to test whether each motor operates normally. Then, a voltmeter is used to check the supply voltage at each interface of the power board to ensure the input voltage to the controller, motors, and other modules is correct. After confirming that the voltage is correct, use a CAN analyzer to check whether the CAN ports of the SPI board and power board are outputting the motor drive signals and verify the stability of the frequency. Finally, if the robot fails to stand in the preset posture (as shown in Figure 7) after startup, please check whether the motor zero positions are set correctly. Due to the varying mechanical characteristics of the robots, the locomotion control parameters need to be adjusted according to the specific circumstances. The robot currently requires manual remote control assistance for docking. Additionally, the robot's docking mechanism utilizes rigid connections, which may not be the optimal choice in certain special cases. In the future, we will explore designs for articulated mechanisms with active degrees of freedom to further enhance the flexibility of modular-legged robots. The integration of a vision module is also among our upcoming research plans, with the aim of enabling autonomous docking based on visual information.

The authors declare that they have no competing financial interests.

The authors would like to express their gratitude to Mr. Xianwu Zeng for his assistance in performing the experiments reported in this paper. This work was supported in part by the National Natural Science Foundation of China (62373223) and the Natural Science Foundation of Shandong Province (ZR2024ZD06).