Reviewer 36 of IROS 2026 submission 2076

Comments to the author
======================

This paper proposes a novel framework for safe robot
navigation in dense crowds. While many existing approaches
rely on the unrealistic assumption that the robot has
direct access to the states of all surrounding humans, this
work simultaneously considers both occlusion and complex
interactions, thereby enabling more realistic and safer
navigation in crowd environments.

The demonstration video is also impressive, and the
proposed approach appears to be a promising step toward
technologies that are essential for robots operating in
real-world environments.

However, real-world environments are significantly more
complex, and there remains substantial room for
improvement. For example, in this work, all humans are
modeled in a unified manner, whereas in real-world
scenarios, diverse types of individuals—such as children,
adults, and elderly people—coexist. This diversity
necessitates more sophisticated prediction models that can
account for heterogeneous human behaviors. In highly
crowded environments such as train stations, such diversity
is particularly prominent. It would therefore be desirable
for future work to extend the proposed framework to handle
such heterogeneous and complex real-world scenarios.

Comments on the Video Attachment
================================

The video is properly prepared.