Table 18 DFNN model performance in full workspace with K-fold.

From: Path-based evaluation of deep learning models for solving inverse kinematics in a revolute-prismatic robot

Configuration

Architecture (hidden layers × units)

Batch size

L2 regularization

Epoch

Validation loss

Single-output (\(\:\theta\:\))

3 × 128

32

0.0005

150

2.1397

Single-output (\(\:d\))

0.09308

Dual-output (\(\:\theta\:,d\))

3 × 128

0.0001

0.3752

Back to article page