Abstract
The analysis of robot inverse kinematic (IK) solutions is fundamental to robot control and path planning. Due to the limitations of analytical and geometric methods, intelligent algorithms are increasingly advantageous as they can derive approximate solutions directly from positive kinematic equations. This paper proposes an improved Particle Swarm Optimization (PSO) algorithm for solving robot inverse kinematics. By integrating an adaptive weight adjustment strategy and a position coefficient $k$ in the velocity factor, the algorithm enhances both global and local search capabilities while reducing computation time. Modeling is performed using the Product of Exponentials (POE) method based on spinor theory, which avoids the singularities inherent in traditional Denavit-Hartenberg (DH) modeling. Experimental results across 6-DOF and 7-DOF robotic arms demonstrate that the proposed algorithm achieves near-zero position error and superior convergence speed compared to standard PSO and Quantum PSO (QPSO) variants.
Introduction
Inverse Kinematics (IK) involves calculating the joint angles required to position a robot's end-effector at a specific target pose. While essential for precision tasks, IK is mathematically complex, often involving non-linear equations with multiple or no solutions. Traditional numerical methods frequently struggle with "singularities"—mathematical configurations where the solver fails—and can be computationally prohibitive for real-time applications.
Intelligent algorithms like PSO offer a robust alternative by treating IK as an optimization problem. However, standard PSO often suffers from slow convergence or becomes trapped in local optima. This research addresses these challenges by introducing an improved PSO framework that combines advanced kinematic modeling with adaptive search strategies.
Kinematic Modeling via Spinor Theory
A core innovation of this research is the adoption of the Product of Exponentials (POE) modeling method based on spinor theory. Unlike the widely used DH parameters, which rely on local coordinate systems for each joint, the POE approach describes the motion of a rigid body as a whole relative to a fixed base frame.
As shown in Figure 1, this method simplifies the description of the robot's geometry. By avoiding local coordinate transformations at every joint, the spinor approach inherently bypasses the mathematical singularities that often plague DH-based solvers, providing a more stable foundation for the optimization process.
Improved Particle Swarm Optimization
The proposed algorithm introduces several key enhancements to the standard PSO architecture to improve its precision and efficiency:
- Adaptive Weight Adjustment: The setting of weights significantly impacts the algorithm's ability to balance global exploration and local exploitation. This paper implements a strategy where weights adjust dynamically based on the current solution's fitness, preventing the particles from overshooting the target or stalling.
- Velocity Factor Optimization: To accelerate computation, a position coefficient $k$ is introduced into the velocity factor, and condition settings based on physical joint limits are applied. This ensures the search remains within the robot's reachable workspace.
- Fitness Function Design: The optimization objective is to minimize the error between the actual end-effector position and the target pose.
As illustrated in Figure 2, the fitness function $f(i)$ calculates the distance between the transformation matrices of the actual and target positions. The goal of the PSO is to drive this distance to zero.
Experimental Validation and Results
The algorithm was tested through three experiments involving a general 6-DOF industrial arm, a PUMA560 arm, and a 7-DOF redundant robotic arm. The proposed method was compared against standard PSO and Quantum PSO (QPSO) across three metrics: accuracy, speed, and convergence.
Precision
The proposed algorithm demonstrated exceptional accuracy. In comparative tests, it achieved a position error of 0 and a maximum orientation error of only $1.29 \times 10^{-8}$ rad. In contrast, the best-performing competing algorithms (PSO and QPSO) showed minimum position errors of $-1.64 \times 10^{-5}$ and orientation errors of $-4.03 \times 10^{-6}$.
Computational Speed
Efficiency is critical for real-time robotics. In the 7-DOF experiment, the proposed algorithm reached a solution in 0.30004 seconds, outperforming the fastest alternative (0.30521s). For the 6-DOF PUMA560, it achieved a time of 0.31851s compared to the next best at 0.33359s.
Convergence Stability
The improved PSO achieved faster and more stable convergence. While standard PSO often required over 350 iterations and exhibited significant fluctuations, the proposed algorithm typically converged within 100 iterations, maintaining its performance even as the complexity of the robotic system increased.
Conclusion
By combining spinor-based POE modeling with an adaptive PSO algorithm, this research provides a high-precision and computationally efficient solution for robot inverse kinematics. The method's ability to maintain accuracy across different robot configurations (6-DOF and 7-DOF) suggests strong potential for multi-arm coordination and complex path planning. Future research will focus on extending this algorithm to real-time trajectory tracking for moving targets.
Summary updated