Answers are grounded in this paper and cite their sources.
Experimental evaluation demonstrates real-time trajectory prediction and collision-aware navigation in dynamic warehouses, suggesting viable scalable automation.
Efficient navigation in dynamic warehouse environments is critical for maintaining productivity and safety within crowded spaces with numerous moving agents. This study addresses the challenge of managing agent movements and ensuring smooth operations by implementing a novel hybrid navigation system that integrates trajectory prediction with collision-aware path planning in response to dynamic agent movements. The system developed in this study introduces a novel approach to navigation by analyzing agent movements through a three-dimensional, spatial–temporal Probabilistic Roadmap graph. It employs a lightweight, custom artificial neural network model that utilizes Long Short-Term Memory and convolutional layers to predict agent trajectories in real time. A unique transformation of location data into relative angle changes, rather than absolute angles, was introduced, yielding improved prediction accuracy. This approach was validated through its application in Warehouse Management Systems, supporting picking-path planning in both simulated and real-world scenarios. The trajectory prediction model achieved an Average Displacement Error of 0.67 m and a Final Displacement Error of 1.01 m in a live setting, with an inference time of less than 1 millisecond. In addition, the proof-of-concept evaluation shows how predicted agent movements can be incorporated into collision-aware path planning. Considering the low computational cost of the trajectory prediction model and the obtained evaluation results, the study provides a viable framework for integrating trajectory prediction with dynamic path planning in warehouse environments.
No takes yet. Share an insight, caveat, or question.
Hering et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: