AI-Driven Inverse Kinematics Calculation for an ABB IRB2400 6-DOF Manipulator Leveraging LSTM and GRU Neural Networks

Inverse kinematics (IK) is a fundamental stage in the design and construction of any robot morphology. It involves calculating the joint values of a robot based on a given initial configuration of its end effector. This process is crucial for enabling the robot to achieve desired positions and orientations in its workspace. Anthropomorphic robots with six degrees of freedom, which mimic human arm movements, present a mathematical complexity that poses a significant challenge for solving inverse kinematics. The complexity arises from the non-linear equations that need to be solved to determine the joint angles from the end effector's position and orientation. Inverse kinematics (IK) calculation presents several challenges, particularly for anthropomorphic robots with six degrees of freedom. Standard inverse kinematic calculation involves solving a set of nonlinear equations that map the desired end effector position and orientation to the corresponding joint angles. This is inherently complex due to the multiple solutions that can exist for a given end effector position and the potential for singularities, where small changes in the end effector position can cause large changes in joint angles. Additionally, traditional analytical methods for solving IK require deriving explicit mathematical expressions, which can be difficult to obtain and computationally expensive to solve, especially for robots with high degrees of freedom. Moreover, numerical methods, while more flexible, can suffer from issues such as convergence to local minima and require good initial guesses to find accurate solutions. These challenges make IK computation a demanding task, particularly for engineers and enthusiasts who may not have a strong mathematical background. In this paper, we address these challenges by employing artificial intelligence techniques, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, to calculate the inverse kinematics of the ABB IRB2400 robot. The dataset used for training and validation is derived from Denavit-Hartenberg-based forward kinematics calculations using 300,000 samples with 20% data for training and 80% for testing. By leveraging these neural intelligent models, we aim to provide a robust and efficient solution to the inverse kinematics problem, reducing the reliance on complex mathematical derivations and