← Back ICRA 2026

Long-Term Human Motion Prediction Using Spatio-Temporal Maps of Dynamics

Yufei Zhu, Andrey Rudenko, Tomasz Piotr Kucner, Achim J. Lilienthal, Martin Magnusson

PDF

AI summary

Key figure (auto-extracted from paper)

A general framework leveraging spatio-temporal Maps of Dynamics outperforms learning-based methods for long-term human motion prediction, with a time-conditioned variant achieving the highest accuracy.

Long-term motion prediction Maps of Dynamics Spatio-temporal modeling Trajectory ranking Human-robot interaction Autonomous navigation

Problem

Accurately predicting human trajectories beyond 20 seconds is critical for robot safety but challenging due to complex environmental influences and time-varying motion patterns that existing methods often overlook.

Approach

MoD-LHMP integrates spatial or spatio-temporal Maps of Dynamics into a velocity-biased prediction model, enhanced with a trajectory ranking method and a novel time-conditioned map to capture daily behavioral variations.

Key results

General MoD-LHMP framework supporting multiple map representations
Up to 50% reduction in average displacement error versus learning-based baselines
Time-conditioned CLiFF-map captures daily motion variations and achieves highest accuracy
Trajectory ranking method enables practical robotic deployment by selecting the most likely path

Why it matters

Enables safer and more reliable autonomous navigation and human-robot interaction in dynamic indoor environments by accurately anticipating long-term pedestrian behavior.

Abstract

Long-term human motion prediction (LHMP) is im- portant for the safe and efficient operation of autonomous robots and vehicles in environments shared with humans. Accurate predictions are important for applications including motion plan- ning, tracking, human-robot interaction, and safety monitoring. In this paper, we exploit Maps of Dynamics (MoDs), which encode spatial or spatio-temporal motion patterns as environment fea- tures, to achieve LHMP for horizons of up to 60 seconds. We pro- pose an MoD-informed LHMP framework that supports various types of MoDs and includes a ranking method to output the most likely predicted trajectory, improving practical utility in robotics. Further, a time-conditioned MoD is introduced to capture motion patterns that vary across different times of day. We evaluate MoD-LHMP instantiated with three types of MoDs. Experiments on two real-world datasets show that MoD-informed method outperforms learning-based ones, with up to 50% improvement in average displacement error, and the time-conditioned variant achieves the highest accuracy overall. Project code is available at https://github.com/test-bai-cpu/LHMP-with-MoDs.git

Index terms

Human Detection and Tracking Autonomous Agents