Researchers Develop Efficient Kinematic Model for Wire-Driven Continuum Robots
In a significant advancement for soft robotics, a team of engineers from Sichuan University and State Grid Ningxia Power Co. has introduced a comprehensive and practical kinematic framework tailored for wire-driven continuum robots. The research, led by Yuanke Chen, presents a novel hybrid methodology that combines the segmented constant curvature approach with a linearly decreasing weight particle swarm optimization (PSO) algorithm, enabling rapid and accurate inverse kinematic solutions for both single- and multi-segment robotic systems. This development addresses a long-standing challenge in the field: the computational complexity and intractability of mapping a desired end-effector position in the workspace back to the required joint and actuator configurations.
Continuum robots, inspired by the flexible bodies of creatures like snakes and elephant trunks, represent a paradigm shift from traditional rigid-link manipulators. Their inherent compliance and dexterity make them uniquely suited for navigating complex, confined, or delicate environments where conventional robots fail. This opens up transformative possibilities in minimally invasive surgery, industrial inspection within machinery, and search-and-rescue operations in collapsed structures. However, the very feature that grants them their advantage—their continuous, flexible backbone—also presents a formidable challenge for control. Unlike rigid robots with well-defined joint angles, the motion of a continuum robot is described by continuous curves, making its kinematic modeling significantly more complex.
The team’s work, published in the peer-reviewed journal Application Research of Computers, focuses on a specific and widely used class of continuum robots: those driven by tendons or wires. In these systems, multiple cables are routed along the robot’s length and anchored at its tip. By differentially pulling these cables, the robot can be made to bend, twist, and extend. The core of the paper’s contribution lies in creating a complete, closed-loop kinematic model that seamlessly connects three distinct spaces: the driving space (the lengths of the actuating wires), the joint space (the geometric parameters like curvature and orientation of each segment), and the workspace (the 3D position and orientation of the robot’s tip).
The researchers began by designing and fabricating a physical prototype of a two-segment wire-driven continuum robot to validate their theoretical framework. This prototype featured two serially connected flexible segments, each actuated by three Nitinol (NiTi) shape-memory alloy wires, providing two degrees of freedom—bending and rotation—per segment. A critical design element was the inclusion of translation segments between and after the actuated joints. These rigid sections, often overlooked in simplified models, play a crucial role in maintaining structural integrity and accurately defining the kinematic chain. The robot’s internal structure was reinforced with a spring to provide a consistent backbone, ensuring predictable bending behavior under actuation.
The kinematic analysis was structured in a systematic, hierarchical manner. The first step involved establishing the mapping between the driving space and the joint space. For a single segment, the relationship between the change in wire length and the resulting curvature and rotation angle can be derived geometrically under the constant curvature assumption, a common and effective simplification. However, the challenge intensifies for multi-segment robots due to inter-segment coupling. When the first segment bends, it alters the base from which the second segment operates. To maintain the second segment’s configuration relative to the first, its driving wires must compensate for the motion of the first segment. The researchers meticulously derived the equations for this coupling effect, calculating the additional wire length changes in the second segment that are induced purely by the movement of the first. This precise accounting for mechanical coupling is a key factor in the model’s accuracy, preventing the significant errors that would arise from treating each segment in isolation.
With the driving-to-joint space mapping established, the next step was to model the forward kinematics: calculating the robot’s tip position in 3D space from the joint angles of both segments. This was achieved using homogeneous transformation matrices, a standard tool in robotics for describing the position and orientation of one coordinate frame relative to another. The researchers defined a series of coordinate frames along the robot’s length—the base frame, the frame at the end of the first segment, the base frame of the second segment, and so on. By multiplying the transformation matrices between each consecutive frame, they obtained the final transformation from the robot’s base to its tip, thereby predicting its end-effector position for any given set of joint angles.
The most computationally challenging part of the problem is the inverse kinematics: given a desired target point in space, what set of joint angles (and consequently, wire lengths) will position the robot’s tip at that point? For a two-segment robot with four degrees of freedom, this is a non-linear, multi-dimensional optimization problem with no closed-form analytical solution. Traditional numerical methods can be slow and prone to getting stuck in local minima, making them unsuitable for real-time control.
To overcome this hurdle, the team employed a Linearly Decreasing Weight Particle Swarm Optimization (LDW-PSO) algorithm. PSO is a bio-inspired evolutionary algorithm that simulates the social behavior of a flock of birds or a school of fish searching for food. A “swarm” of potential solutions, called “particles,” is initialized with random positions and velocities in the search space (in this case, the space of possible joint angles). Each particle remembers its own best-known position (personal best) and is aware of the best-known position found by any particle in the swarm (global best). In each iteration, the particles update their velocity and position based on these two pieces of information, gradually converging toward an optimal solution.
The “linearly decreasing weight” modification is crucial for balancing exploration and exploitation. The inertia weight controls how much of a particle’s previous velocity is retained. A high initial weight allows the swarm to explore a wide area of the search space, preventing premature convergence. As the algorithm progresses, this weight is linearly decreased, encouraging the particles to fine-tune their search and converge more precisely on the global optimum. The researchers defined the algorithm’s fitness function as the Euclidean distance between the robot’s current tip position (calculated via forward kinematics) and the desired target position. The goal of the PSO is to minimize this distance to zero.
The study’s strength lies in its thorough validation through both simulation and physical experimentation. In simulation, the researchers tested the algorithm’s performance by setting a target point and using the LDW-PSO to find the necessary joint angles. The results demonstrated rapid convergence, with the position error decreasing to an acceptable level within a small number of iterations. To rigorously test the algorithm’s speed, a series of timing experiments were conducted. The researchers performed single-point inverse kinematic calculations ten times and multi-point calculations (1,000 points) ten times. The average solving time for a single point was a mere 0.019 seconds, and for multiple points, it was an even more impressive 0.012 seconds. This exceptional speed is a critical achievement, as it makes real-time, closed-loop control of the robot a practical reality.
The ultimate test of the model’s validity was conducted on the physical robot prototype. The experimental setup was sophisticated, featuring a high-precision optical motion capture system (Nokov) with eight infrared cameras. Seven retro-reflective markers were strategically placed along the robot’s body—on the base, at the joints, in the middle of each segment, and on the tip. This system could track the 3D position of these markers with a claimed spatial accuracy of ±0.1 mm, providing a highly accurate measurement of the robot’s actual shape and tip position.
The experimental protocol was methodical. Ten arbitrary target points were selected within the robot’s reachable workspace. For each point, the inverse kinematics algorithm was used to compute the required wire length changes. An STM32-based embedded controller then commanded the six DC motors to adjust the wire lengths accordingly. After the robot settled at the target, the optical system recorded the actual tip position.
The results were highly encouraging. The average position error across all ten target points was measured to be less than 18.9 mm. When normalized by the robot’s total body length, this corresponds to an average error of 6.22%. The researchers broke this down by axis, finding average errors of 8.89 mm (2.92% of body length) in the x-axis, 5.95 mm (1.96%) in the y-axis, and 5.67 mm (1.87%) in the z-axis. This level of accuracy is considered very good for a soft robot, especially one operating in a gravity field.
The paper also includes a compelling visual comparison between the robot’s theoretical arm shape (predicted by the model) and its actual arm shape (reconstructed from the marker positions). The close alignment between the two curves provides strong qualitative evidence for the model’s fidelity. The researchers attributed the residual error to several practical factors: the gravitational sag of the robot’s body, manufacturing tolerances in the physical components, and minor inaccuracies inherent in the optical tracking system.
This research represents a significant step forward in the practical deployment of continuum robots. By providing a complete, fast, and experimentally validated kinematic model, the team has created a foundational tool that can be readily applied to a wide range of similar robotic systems. The integration of the PSO algorithm for inverse kinematics is particularly elegant, as it is both robust and computationally efficient. The inclusion of translation segments and the careful handling of inter-segment coupling elevate the model from a theoretical exercise to a practical engineering solution.
The implications of this work are broad. In the medical field, such a model could enable a surgeon to command a robotic endoscope to navigate to a precise location within the human body with a simple point-and-click interface, with the underlying algorithm handling the complex wire actuation. In industrial settings, a robot could be programmed to inspect the interior of a jet engine or a pipeline with high precision. The speed of the algorithm means that the robot can respond dynamically to changes in its environment or task.
Future work, as mentioned by the authors, will focus on incorporating environmental constraints, such as obstacles, into the kinematic planning. This would involve using the fast inverse kinematics solver as a core component within a higher-level path-planning algorithm, allowing the robot to not only reach a point but to do so while avoiding collisions. The success of this current work provides a solid platform upon which to build these more advanced capabilities, bringing the promise of highly dexterous, soft robotic systems closer to widespread application.
Chen Yuanke, Ma Feiyue, Xiang Guofei, Ma Congjun, Chen Lei, Ni Hui, Dian Songyi. Sichuan University and State Grid Ningxia Power Co. Application Research of Computers. DOI: 10.19734/j.issn.1001-3695.2021.03.0060