An Enhanced AI-Based Approach for Contextualization and Prediction of Long-Term Human Activities in Industrial Robot Applications
Sebastian Krusche, Jayanto Halim, Shuxiao Hou, Mohamad Bdiwi, Steffen Ihlenfeldt · 2026
Abstract Predicting long-term human activities and movement patterns in industrial environments with content semantics using RGB color images and 3D point clouds can enhance the efficiency and effectiveness of human–robot collaboration applications. This work aims to develop methods for predicting complex, long-term human activities and estimating movements within industrial processes. The main contributions of this work are: 1. Designing a multi-layered structure with various methods for predicting long-term human intentions and walking paths in an industrial context, 2. Conducting empirical experiments with 60 subjects to collect data on human actions and activities in an industrial environment, and 3. Training a comprehensive end-to-end multitask learning system for long-term prediction of human activities and movement patterns. These approaches were integrated into the proposed framework and evaluated across six scenarios in an industrial use case.