How Human Motion Prediction Quality Shapes Social Robot Navigation Performance in Constrained Spaces

Andrew Stratton1, Phani Teja Singamaneni2, 3, Pranav Goyal1, Rachid Alami2, Christoforos Mavrogiannis1,
1University of Michigan, 2LAAS-CNRS, 3INRIA
Paper arXiv Video Code

Abstract

Motivated by the vision of integrating mobile robots closer to humans in warehouses, hospitals, manufacturing plants, and the home, we focus on robot navigation in dynamic and spatially constrained environments. Ensuring human safety, comfort, and efficiency in such settings requires that robots are endowed with a model of how humans move around them. Human motion prediction around robots is especially challenging due to the stochasticity of human behavior, differences in user preferences, and data scarcity. In this work, we perform a methodical investigation of the effects of human motion prediction quality on robot navigation performance, as well as human productivity and impressions. We design a scenario involving robot navigation among two human subjects in a constrained workspace and instantiate it in a user study (N=80) involving two different robot platforms, conducted across two sites from different world regions. Key findings include evidence that: 1) the widely adopted average displacement error is not a reliable predictor of robot navigation performance and human impressions; 2) the common assumption of human cooperation breaks down in constrained environments, with users often not reciprocating robot cooperation, and causing performance degradations; 3) more efficient robot navigation often comes at the expense of human efficiency and comfort.

Video

Experimental Setup

Our experimental setup is modeled after constrained, indoor domains of interest such as hospitals, homes, and warehouses. Participants cross the workspace picking and stacking blocks at different workstations, while the robot moves between them. The structure facilitates consistent, multihuman-robot interactions in the shared space.

Experimental workspace setup

To enhance the applicability of our findings, our study was run at both the University of Michigan in the USA and LAAS-CNRS at the University of Toulouse in France, with distinct robots at each site (the Hello Robot Stretch 2 at UM and Willow Garage PR2 at LAAS).

University of Michigan experimental setup

UM

LAAS-CNRS experimental setup

LAAS

Findings

Our primary finding is that, generally, Average Displacement Error (ADE) does not correlate with measures of Social Robot Navigation performance. Specifically, while it does positively correlate with users' Average Accelerations, it has no correlation with their self-reported Discomfort or perceived workload, and is actually negatively correlated with their Path Irregularity and the team's Goals Per Second (GPS).

ADE correlation with average accelerations
ADE correlation with discomfort
ADE correlation with team goals per second
ADE correlation with path irregularity

Additionally, we find that our chosen transformer and optimization based predictors do not offer improved Social Robot Navigation performance over simpler models which treat humans as generic static or dynamic obstacles.

Average acceleration by prediction algorithm
Discomfort by prediction algorithm
Frustration by prediction algorithm
Path irregularity by prediction algorithm

We also perform several additional exploratory analyses in the paper.

BibTeX

@article{stratton2026pred2nav,
  author    = {Stratton, Andrew and Singamaneni, Phani Teja and Goyal, Pranav and Alami, Rachid and Mavrogiannis, Christoforos},
  title     = {How Human Motion Prediction Quality Shapes Social Robot Navigation Performance in Constrained Spaces},
  journal   = {Proceedings of the ACM/IEEE International Conference on Human Robot Interaction (HRI)},
  year      = {2026},
}