Sign in Subscribe
Robotics industry

Deep Learning Models for Solving Inverse Kinematics in Robotic Manipulators

Strategic Insights

— 1 min read

Researchers at Nitte (Deemed to be University) have explored the effectiveness of deep learning models in solving the inverse kinematics problem in robotic manipulators. The study aimed to evaluate the performance of different neural architectures in predicting joint configurations from end-effector positions across various workspace regions. The investigation employed two training strategies, quadrant-based and full workspace training, and applied k-fold cross-validation to improve robustness.

Key Takeaways:

  • The deep feedforward neural network (DFNN) with k-fold cross-validation (CV) achieved the lowest Cartesian deviation errors, with an average error of 0.531 mm in the full workspace.
  • The DFNN with k-fold CV consistently outperformed other models, including DFNN without k-fold CV, long short-term memory (LSTM), and gated recurrent unit (GRU).
  • The study demonstrated that the k-fold CV-based DFNN with a single-output formulation is capable of handling singularities and ambiguity in joint solutions.
  • The research highlighted the importance of using a suitable training strategy and robustness techniques, such as k-fold CV, to improve the performance of deep learning models.
  • The findings of this study have implications for the development of more advanced and reliable robotic manipulators.

Statistics:

  • The average error of the DFNN with k-fold CV in the full workspace was 1.594 mm (Square) and 2.084 mm (Circle).
  • The Cartesian deviation errors for the DFNN with k-fold CV in the quadrants were 0.289 mm in Q1, 0.410 mm in Q2, 0.508 mm in Q3, and 0.715 mm in Q4 on the Square path.
  • The errors for the DFNN with k-fold CV on the Circle path in the quadrants were 0.312 mm, 0.366 mm, 0.438 mm, and 0.662 mm in Q1 to Q4 respectively.

Sources:

  • "Path-based evaluation of deep learning models for solving inverse kinematics in a revolute-prismatic robot" (Scientific Reports, 2025,15(1):1-17)
  • "Studies from Nitte (Deemed to be University) Update Current Data on Robotics" (Journal of Engineering, October 20, 2025, p 3999)
  • Nitte (Deemed to be University)
  • Navya Manjegowda, Department of Mathematics, Nitte (Deemed to be University), NMAM Institute of Technology (NMAMIT)
  • Muralidhara Rao

Read more