Year: 2025
Type: article
Abstract: This research explores the intersection of artificial intelligence and robotics, focusing on integrating machine learning for predictive modeling in robotic systems. Specifically, it explores the use of four regression algorithms, decision tree, random forest, Ada boost, and XG boost, to predict battery voltage based on key operational parameters such as joint angles, motor torque, sensor latency, and end-of-performance error. A dataset of 20 observations was used to analyze performance and find relationships between mechanical factors and power efficiency. The analysis found strong correlations between end-of-performance error and motor torque (0.97), joint angle 1 (0.93), and sensor latency (0.89), while showing a strong negative correlation with battery voltage (-0.95). The performance evaluation revealed distinct patterns across the models. Decision tree regression achieved perfect accuracy on the training data (R² = 1.0000) but underperformed on the test data (R² = 0.8040), indicating overfitting. Random Forest provided the best overall performance, with a strong training score (R² = 0.9742) and the highest generalization on the test data (R² = 0.8553). AdaBoost performed well on training (R² = 0.9986) but showed moderate test accuracy (R² = 0.7666). XG Boost matched the decision tree on training accuracy (R² = 1.0000) but struggled with generalization (R² = 0.6570), indicating significant overfitting. This study underscores the critical role of battery voltage monitoring in robotic systems, as power fluctuations directly affect position accuracy and overall system reliability. The findings suggest that ensemble models, especially Random Forest, offer a more stable balance between accuracy and generalization, making them suitable for real-time predictive maintenance and system optimization in robotics.
Source: CompSci & AI Advances
Authors: Abdul Samad, Mohammad Jainul
Institution: Vellore Institute of Technology University
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7204276366
FWCI:
Cites:
Cited by:
Related to: