Disentangling reactive control, representation, map knowledge and prediction in robot navigation among pedestrians: A controlled simulation benchmark

Benchmarks of robot navigation among pedestrians usually compare complete planners, so the design choice behind a difference is rarely identified. We present a seeded, paired simulation benchmark that varies the planning representation (sparse Voronoi roadmap, dense lattice, dense grid), a local reactive potential field (on or off), the robot’s knowledge (full, or from a 120° occluded sensor without a map) and pedestrian forecasting (from constant velocity to a trained long short-term memory and an oracle). Each operating point uses 100 paired seeds, analysed with multiplicity-adjusted paired tests and with failure-inclusive and time-to-collision metrics. The reactive field is the dominant lever: it raises minimum pedestrian clearance by 0.31–1.18 m for every representation and cuts the share of runs entering a 0.9-m contact zone from up to 69% to at most 4%; with it, the roadmap and lattice differ by at most 0.06 m. Forecasting and map completeness matter little: forecasts add about 0.1 m at most, perfect forecasts from an oracle perform like constant velocity, a full static map changes nothing, and omniscient pedestrian tracking adds 0.09–0.18 m. Reliability differences between representations vanish once the roadmap is queried correctly; we document this and three further artefacts – planner-coupled random streams, an unstated robot-to-pedestrian speed ratio and non-inflated obstacles in external planner ports – that can manufacture effects of the size typically reported. The orderings survive sensor noise, social-force and ORCA crowds, an indoor scene family and multiple robots, but weaken when the robot moves no faster than pedestrians under unicycle limits.

Authors

Institutions

Publication Details

Journal
SIMULATION
Published
2026-11-20
DOI
https://doi.org/10.1177/00375497261508045
Primary Topic
Robotic Path Planning Algorithms
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Disentangling reactive control, representation, map knowledge and prediction in robot navigation among pedestrians: A controlled simulation benchmark

Yao‐Jen Chang, Hung-Huan Liu
SIMULATION
Robotic Path Planning Algorithms
article

Disentangling reactive control, representation, map knowledge and prediction in robot navigation among pedestrians: A controlled simulation benchmark

Yao‐Jen Chang, Hung-Huan Liu
article en

Abstract

Benchmarks of robot navigation among pedestrians usually compare complete planners, so the design choice behind a difference is rarely identified. We present a seeded, paired simulation benchmark that varies the planning representation (sparse Voronoi roadmap, dense lattice, dense grid), a local reactive potential field (on or off), the robot’s knowledge (full, or from a 120° occluded sensor without a map) and pedestrian forecasting (from constant velocity to a trained long short-term memory and an oracle). Each operating point uses 100 paired seeds, analysed with multiplicity-adjusted paired tests and with failure-inclusive and time-to-collision metrics. The reactive field is the dominant lever: it raises minimum pedestrian clearance by 0.31–1.18 m for every representation and cuts the share of runs entering a 0.9-m contact zone from up to 69% to at most 4%; with it, the roadmap and lattice differ by at most 0.06 m. Forecasting and map completeness matter little: forecasts add about 0.1 m at most, perfect forecasts from an oracle perform like constant velocity, a full static map changes nothing, and omniscient pedestrian tracking adds 0.09–0.18 m. Reliability differences between representations vanish once the roadmap is queried correctly; we document this and three further artefacts – planner-coupled random streams, an unstated robot-to-pedestrian speed ratio and non-inflated obstacles in external planner ports – that can manufacture effects of the size typically reported. The orderings survive sensor noise, social-force and ORCA crowds, an indoor scene family and multiple robots, but weaken when the robot moves no faster than pedestrians under unicycle limits.

SIMULATION
Chung Yuan Christian University (TW)
Robotic Path Planning Algorithms
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Disentangling reactive control, representation, map knowledge and prediction in robot navigation among pedestrians: A controlled simulation benchmark — Yao‐Jen Chang, Hung-Huan Liu · SIMULATION (2026) | TGRS Research Map | TGRS