🌐 English · 한국어
Summary
This paper reports a real-robot field experiment on passive pedestrian-robot interaction (HRI) in an exit corridor. A mobile robot oscillates back and forth in a direction perpendicular to a uni-directional pedestrian flow heading toward an exit. Pedestrians see the robot and avoid it by adjusting their trajectory or speed, passively interacting with it. Across 6 cases (normal vs. faster walking speed; no robot / slow robot / fast robot), the authors show that the robot’s presence lowers the crowd’s average speed, and that a faster robot lowers it further. The results qualitatively confirm the authors’ earlier social-force simulations, supporting the use of robots to regulate pedestrian flow (e.g., to counter the faster-is-slower effect in evacuations).
Key Contributions
- First real-robot empirical validation (beyond simulation) of robot-assisted pedestrian flow regulation in an exit corridor.
- Demonstrates a monotonic slow-down effect: robot present lowers average pedestrian velocity, and faster robot motion lowers it more.
- Shows the effect is stronger for faster-walking pedestrian groups.
- Establishes qualitative agreement between field experiments and prior social-force-model simulations.
Methodology and Architecture
- Setup: 10 m x 4 m tracked corridor with an exit door, at Stevens Institute of Technology (June 22, 2016).
- Tracking: 5 Microsoft Kinect sensors + networked computers running OpenPTrack (RGB-D people tracking on ROS); checkerboard-based extrinsic calibration and a homogeneous transform (solved over 55 grid vertices) mapping to the ground frame.
- Robot: Customized Adept Pioneer P3-DX, single-integrator dynamics. Velocity perpendicular to the flow follows simple harmonic motion u_r^y = Aω sin(ωt), with amplitude A = 2 m (half corridor width); x-velocity = 0. Start position (6.6, 0).
- Design: 6 cases x 11 naive participants x 3 runs. Two walking speeds (normal: Cases 1-3; faster: Cases 4-6), each with three robot conditions — no robot, slow robot (ω = 0.1 rad/s), fast robot (ω = 0.4 rad/s).
- Analysis: Velocities from backward difference + moving-average filter; average velocity profiles vs. x-position; box plots of velocity distributions; flow velocity averaged within an “HRI region” (within 2 m of the robot path) for comparison with simulation.
Results
- Individual: Pedestrians change direction (around x ≈ 4.5-4.8 m) to avoid anticipated collisions, sometimes also slowing down; both trajectory and velocity are affected.
- Collective: Average pedestrian velocity is lowered with the robot present; faster robot → lower velocity. The effect is concentrated in the HRI region around the robot’s path. Box-plot quartiles, medians, and third quartiles all show the same trend.
- Speed dependence: The slow-down is more significant for the faster-moving pedestrian group.
- Validation: Qualitative agreement with the authors’ earlier social-force simulations [Jiang et al. 2016]; three core claims (speed affected by robot; faster robot slows flow more; effect stronger at higher pedestrian speed) hold in both. No quantitative/statistical significance tests reported.
- Limitations: Small, single-density dataset; qualitative agreement only; assumes rational, non-panic, clear-visibility conditions. Future work: tune simulations with the data and build a learning-based motion planner to regulate collective speed.
Related Papers
- gehrke-2022-evaluation-of-sidewalk-autonomous-delivery — Both are field studies of robots operating among pedestrians; Gehrke evaluates a sidewalk delivery robot in outdoor pedestrian space, whereas this paper studies how a robot intentionally regulates indoor crowd flow — complementary views of robot impact on pedestrian movement.
- rouphail-1998-recommended-procedures-for-chapter-13 — Provides pedestrian flow/level-of-service and capacity methodology; relevant background for quantifying the average-speed and density effects that this robot-regulation experiment measures and manipulates.