Neural Implicit Flow Fields for Spatio-Temporal Motion Mapping

2025-10-16

AI Summary

Robot manipulation in complex human environments demands simultaneous safety and efficiency, yet existing motion-oriented dynamic maps (MoDs) rely on discrete spatial sampling, incur high construction costs, and lack online update capability. Method: This paper proposes a continuous spatiotemporal dynamic graph representation based on implicit neural functions. It directly maps spatiotemporal coordinates to parameters of a semi-wrapped Gaussian mixture model, integrating continuous spatiotemporal positional encoding with probabilistic motion representation—enabling smooth extrapolation and online construction without grid-based sampling. Contribution/Results: Evaluated on a real-world long-term pedestrian trajectory dataset, our method significantly improves motion pattern modeling accuracy and velocity distribution smoothness over baseline approaches, while maintaining efficient inference. It establishes a scalable, lightweight paradigm for motion prior modeling in dynamic robotic navigation.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Safe and efficient robot operation in complex human environments can benefit from good models of site-specific motion patterns. Maps of Dynamics (MoDs) provide such models by encoding statistical motion patterns in a map, but existing representations use discrete spatial sampling and typically require costly offline construction. We propose a continuous spatio-temporal MoD representation based on implicit neural functions that directly map coordinates to the parameters of a Semi-Wrapped Gaussian Mixture Model. This removes the need for discretization and imputation for unevenly sampled regions, enabling smooth generalization across both space and time. Evaluated on a large public dataset with long-term real-world people tracking data, our method achieves better accuracy of motion representation and smoother velocity distributions in sparse regions while still being computationally efficient, compared to available baselines. The proposed approach demonstrates a powerful and efficient way of modeling complex human motion patterns.

Problem

• Modeling site-specific human motion patterns for robot navigation
• Creating continuous spatio-temporal motion maps without discretization
• Improving motion representation accuracy in sparse tracking data regions

Innovation

• Uses implicit neural functions for continuous motion mapping
• Directly maps coordinates to Semi-Wrapped Gaussian parameters
• Enables smooth spatio-temporal generalization without discretization