Robot Intention Recognition for Near Future Collision Avoidance.

Author

Naga Sai Rishik Reddy Vaka, Southern Illinois University CarbondaleFollow

Date of Award

5-1-2025

Degree Name

Master of Science

Department

Computer Science

First Advisor

Henry, Hexmoor

Abstract

AN ABSTRACT OF THE THESIS OFVaka Naga Sai Rishik Reddy, for the Master of Science degree in Computer Science, Presented on March 2025, at Sothern Illinois University Carbondale. TITLE: Robot Intention Recognition for Near Future Collision Avoidance. MAJOR PROFESSOR: Dr. Henry Hexmoor, Ph.D. This case study introduces a novel deep-learning approach for real-time collision avoidance in dynamic environments. Traditional models, such as RVO2, ORCA, Social-LSTM, and SocialGAN, have demonstrated limitations in multi-agent interaction prediction, real-time adaptability, and uncertainty estimation. The proposed Adaptive Intent Prediction Model (AIPM) integrates Transformers, Bayesian Inference, and Meta-Learning to overcome these challenges. AIPM leverages sensor data to predict future trajectories, estimate uncertainties, and dynamically adapt to agent behaviors. Through comparative evaluation against traditional methods, AIPM demonstrates superior prediction accuracy, collision avoidance capabilities, and computational efficiency. This study provides a structured implementation of AIPM along with dataset preprocessing, model training, and evaluation methodologies. The findings highlight AIPM’s potential for enhancing autonomous robot and vehicle safety in real-world scenarios.