Inverse kinematics using neural networks and random forests for trajectory tracking of a three-degree-of-freedom robotic arm
| dc.audience | companies | |
| dc.audience | students | |
| dc.audience | researchers | |
| dc.audience | teachers | |
| dc.audience | generalPublic | |
| dc.contributor.author | Ramírez-Vasquez, Jose Francisco | |
| dc.contributor.author | Carlock-Acevedo, Emanuel de Jesus | |
| dc.contributor.author | Castro-Ramírez, Joel | |
| dc.date.accessioned | 2024-09-18T06:28:29Z | |
| dc.date.available | 2024-09-18T06:28:29Z | |
| dc.date.issued | 2024-09-14 | |
| dc.description.abstract | Direct and inverse kinematics are crucial in the operation of manipulator robots to achieve desired positions and orientations and execute specific tasks through precise trajectory tracking in the workspace. This study addresses the challenge of inverse kinematics for a three-degree-of-freedom robot, highlighting its complexity due to the nonlinear nature of the trigonometric equations and the existence of multiple solutions for a given end-effector position. To solve this problem, two machine learning approaches were implemented: artificial neural networks and random forests. First, the direct kinematics model was obtained using the Denavit-Hartenberg method. With these equations, a training dataset was generated by positioning the robot at various points within the workspace using MATLAB and the Robotics Toolbox by Peter Corke. The models were developed in Python using TensorFlow, Keras, and Scikit-learn. The neural network was adjusted by increasing the number of neurons in the hidden layers until a satisfactory response was obtained, while the random forest model was optimized by varying the number of decision trees. Both models were evaluated with previously unseen trajectories, and their performance was compared using graphs generated in Python and trajectory tracking simulations in MATLAB. The results demonstrated that both approaches can effectively adapt to multidimensional regression problems when provided with an appropriate dataset. In this context, the inverse kinematics problem for a three-degree-of-freedom robotic manipulator was posed as a regression problem, using neural network and random forest models to map a set of numerical inputs to a set of numerical targets, thereby demonstrating the feasibility of these approaches in solving complex robotics problems. | |
| dc.folio | AyTBUAP 9(EIA):1-23 | |
| dc.format | ||
| dc.identificator | 7 | |
| dc.identifier.uri | https://hdl.handle.net/20.500.12371/21309 | |
| dc.language.iso | spa | |
| dc.publisher | Benemérita Universidad Autónoma de Puebla | |
| dc.rights.acces | openAccess | |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc-sa/4.0 | |
| dc.subject.classification | INGENIERÍA Y TECNOLOGÍA | |
| dc.title | Inverse kinematics using neural networks and random forests for trajectory tracking of a three-degree-of-freedom robotic arm | |
| dc.type | Artículo | |
| dc.type.conacyt | article |
