On solving the inverse kinematics problem using neural networks

Csiszar, Akos; Eilers, Jan; Verl, Alexander

doi:10.1109/m2vip.2017.8211457

Cited by 51 publications

(32 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The notion of utilizing artificial neural networks for inverse kinematics and robot control was explored back in 1993 (Jack et al, 1993 ) and more recently revisited by Csiszar et al ( 2017 ). Neuromorphic implementations, which are based on SNN, have gained tremendous traction in past decay due to the increased attention to neurorobotics and, more recently, the emergent availability of neuromorphic software and hardware frameworks.…”

Section: Discussionmentioning

confidence: 99%

Neuromorphic NEF-Based Inverse Kinematics and PID Control

et al. 2021

View full text Add to dashboard Cite

Neuromorphic implementation of robotic control has been shown to outperform conventional control paradigms in terms of robustness to perturbations and adaptation to varying conditions. Two main ingredients of robotics are inverse kinematic and Proportional–Integral–Derivative (PID) control. Inverse kinematics is used to compute an appropriate state in a robot's configuration space, given a target position in task space. PID control applies responsive correction signals to a robot's actuators, allowing it to reach its target accurately. The Neural Engineering Framework (NEF) offers a theoretical framework for a neuromorphic encoding of mathematical constructs with spiking neurons for the implementation of functional large-scale neural networks. In this work, we developed NEF-based neuromorphic algorithms for inverse kinematics and PID control, which we used to manipulate 6 degrees of freedom robotic arm. We used online learning for inverse kinematics and signal integration and differentiation for PID, offering high performing and energy-efficient neuromorphic control. Algorithms were evaluated in simulation as well as on Intel's Loihi neuromorphic hardware.

show abstract

Section: Discussionmentioning

confidence: 99%

Neuromorphic NEF-Based Inverse Kinematics and PID Control

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Among those ANN techniques, Multi-Layer Perceptron (MLP) has been extensively used for task-driven IK learning problems [11][12][13][14][15]. Besides limiting the problem to task-driven workspaces, early systems also considered only a small number of DoF.…”

Section: Survey On Data-driven Methodsmentioning

confidence: 99%

“…The authors observed that the MLP employed provided minimum mean square error for a dataset of 1000 data points. In a more recent system, Csiszar et al [13] used two MLP networks, a (3/50/50/4) and a (4/50/50/6), to learn the inverse kinematics of 3DoF and 4DoF serial robots respectively in a task-independent workspace. They generated a much larger dataset with 25000 data points for the 3DoF and 75000 data points for the 4DoF, but they constrained the workspaces by limiting the ranges of the joints of the robots.…”

Section: Survey On Data-driven Methodsmentioning

confidence: 99%

“…For IK of task-driven workspaces, ANN methods have been successful in learning and predicting data points lying on specific trajectories (planar, circular, spring shapes, etc.) [11][12][13][14][15]. However, not much effort have been dedicated to investigating solutions for IK of task-independent workspaces.…”

Section: Introductionmentioning

confidence: 99%

“…13: Results for 5 DoF robot using the structured dataset. (6.350e-12/5.738e-5)(1.369e-13/1.097e-4) (0.000e+0/5.108e-6)…”

mentioning

confidence: 99%

See 2 more Smart Citations

Use of Jacobians for inverse kinematics of articulated robots : a study on approximate solutions

Demby's¹

View full text Add to dashboard Cite

In the context of articulated robotic manipulators, the Forward Kinematics (FK) is a highly non-linear function that maps joint configurations of the robot to poses of its endeffector. Furthermore, while in the most useful cases these functions are neither injective (one-to-one) nor surjective (onto), depending on the robot configuration -- i.e. the sequence of prismatic versus revolute joints, and the number of Degrees of Freedom (DoF) -- the associated Inverse Kinematics (IK) problem may be practically or even theoretically impossible to be solved analytically. Therefore, in the past decades, several approximate methods have been developed for many instances of IK problems. The approximate methods can be divided into two distinct categories: data-driven and numerical approaches. In the first case, data-driven approaches have been successfully used for small workspace domains (e.g., task-driven applications), but not fully explored for large ones, i.e. in task-independent applications where a more general IK is required. Similarly, and despite many successful implementations over the years, numerical solutions may fail if an improper matrix inverse is employed (e.g., Moore-Penrose generalized inverse). In this research, we propose a systematic, robust and accurate numerical solution for the IK problem using the Unit-Consistent (UC) and the Mixed (MX) Inverse methods to invert the Jacobians derived from the Denavit-Hartenberg (D-H) representation of the FK for any robot. As we demonstrate, this approach is robust to whether the system is underdetermined (less than 6 DoF) or overdetermined (more than 6 DoF). We compare the proposed numerical solution to data driven solutions using different robots -- with DoF varying from 3 to 7. We conclude that numerical solutions are easier to implement, faster, and more accurate than most data-driven approaches in the literature, specially for large workspaces as in task-independent applications. We particularly compared the proposed numerical approach against two data-driven approaches: Multi-Layer Perceptron (MLP) and Adaptive Neuro-Fuzzy Inference System (ANFIS), while exploring various architectures of these Neural Networks (NN): i.e. number of inputs, number of outputs, depth, and number of nodes in the hidden layers.

show abstract

Optimizing cotton‐picking robotic manipulator and inverse kinematics modeling using evolutionary algorithm‐assisted artificial neural network

Singh,

Tewari,

Biswas

et al. 2023

Journal of Field Robotics

View full text Add to dashboard Cite

This study presents a particle swarm optimization (PSO) algorithm‐assisted neural‐network‐based inverse kinematics solution for a 4‐DoF (degree‐of‐freedom) cotton harvesting robot. A novel setup was developed to measure the three‐dimensional locations of in‐field cotton bolls. Dimensional optimization of the manipulator was conducted using the PSO algorithm to minimize torque requirements at joints. With the optimized links’ lengths, the targeted end‐effector positions were achieved effectively (coefficient of determination (R2) > 99.88). The genetic algorithm optimized the neural network architecture to include three hidden layers with [64 64 32] neurons, identifying the Tanh activation function as the optimal configuration. A custom loss function was used during the training of artificial neural network (ANN). Using angles predicted by the trained ANN, the end‐effector reached targeted positions with positioning errors below 13.0 mm. A hybrid model consisting of an ANN and PSO algorithm was developed to further reduce the error. This trained hybrid model resulted in a positioning error below 1.0 mm with inference time of 6.07 s during simulation phase. As compared to the ANN and PSO algorithm, hybrid model reduced the positioning error and inference time (>40.0%), respectively. For hybrid model, the mean percentage errors of 0.25%, 0.39%, and 0.84% were observed along the x‐, y‐, and z‐axis. A positioning error below 9.0 mm occurred during evaluation of the hybrid model with the fabricated manipulator. Hence, the developed hybrid model precisely determines the joint angles, allowing the end‐effector of the cotton harvesting robot to reach at targeted pose with minimum error.

show abstract

On solving the inverse kinematics problem using neural networks

Cited by 51 publications

References 8 publications

Neuromorphic NEF-Based Inverse Kinematics and PID Control

Neuromorphic NEF-Based Inverse Kinematics and PID Control

Use of Jacobians for inverse kinematics of articulated robots : a study on approximate solutions

Optimizing cotton‐picking robotic manipulator and inverse kinematics modeling using evolutionary algorithm‐assisted artificial neural network

Contact Info

Product

Resources

About