Follow-Bench: A Unified Motion Planning Benchmark for Socially-Aware Robot Person Following

TL;DR

This work addresses the challenge of safe and comfortable robot person following (RPF) in dynamic, human-populated environments. It provides a structured literature review, introduces Follow-Bench a unified 2D motion-planning benchmark, and systematically re-implements six RPF planners to evaluate safety-comfort trade-offs across diverse scenarios. Key findings show model-predictive control (MPC)-based planners, especially MPC w/ Traj., offer the strongest performance by leveraging target trajectory forecasts, improving target visibility and collision avoidance, though high-density scenarios still degrade performance and side-following is particularly challenging. Real-world experiments on a differential-drive robot corroborate the simulation results, underscoring the potential and limitations of current RPF planners and highlighting avenues for future research in adaptive following, occlusion handling, and richer environmental representations.

Abstract

Robot person following (RPF) -- mobile robots that follow and assist a specific person -- has emerging applications in personal assistance, security patrols, eldercare, and logistics. To be effective, such robots must follow the target while ensuring safety and comfort for both the target and surrounding people. In this work, we present the first comprehensive study of RPF, which (i) surveys representative scenarios, motion-planning methods, and evaluation metrics with a focus on safety and comfort; (ii) introduces Follow-Bench, a unified benchmark simulating diverse scenarios, including various target trajectory patterns, crowd dynamics, and environmental layouts; and (iii) re-implements six representative RPF planners, ensuring that both safety and comfort are systematically considered. Moreover, we evaluate the two best-performing planners from our benchmark on a differential-drive robot to provide insights into the real-world deployment of RPF planners. Extensive simulation and real-world experiments provide a quantitative study of the safety-comfort trade-offs of existing planners, while revealing open challenges and future research directions.

Figures (15)

• Figure 1: Social space definition. A desired following position is within the personal zone (0.45-1.2 meters) of the target, while avoiding the private zone (less than 0.45 meters) of surrounding pedestrians. The distance is measured as $|d_{rh}-R_h-R_r|$, where $d_{rh}$ is the distance between the robot and the target, $R_h$ is the radius of the target, and $R_r$ is the radius of the robot.
• Figure 2: A general motion-planning framework for RPF emphasizing safety and comfort constraints (e.g., proxemics, collision avoidance, target visibility). The global planner uses an estimated long-term goal of the target person to generate a global path (optional in most existing methods), while the local planner is essential for producing local trajectories to follow the target person, ensuring safety and comfort concerns. In this paper, we mainly review and benchmark RPF-oriented local motion planners.
• Figure 3: Scenarios of target trajectories. The scenarios include two-triangle, two-square, U-turn, figure-eight ('8'), L-turns with varying angles (30$^{\circ}$, 45$^{\circ}$, and 60$^{\circ}$), and back-and-forth walking. These scenarios are used to independently evaluate planners' agility and motion smoothness. The robot is initialized at its preferred following position relative to the target including back-following and side-following. The green rectangle represents the robot, the orange circle represents the target, and the short red curve denotes the desired following trajectory generated by the planner. The long red lines around the robot are simulated 2D LiDAR scans.
• Figure 4: Scenarios of dynamic crowds. Blue circles denote dynamic pedestrians, and the orange circle denotes the target. In each scenario, the number of pedestrians is varied from 5 to 30 to create different crowd densities, enabling independent evaluation of planners' collision-avoidance and social-spacing capabilities in human-populated environments. The scenarios include circular crowds (flow around a central area, e.g., courtyards), random crowds (e.g., plazas or shopping malls), parallel crossing (e.g., pedestrians in crosswalks), and perpendicular crossing (e.g., intersections in open spaces).
• Figure 5: Scenarios of layouts with structured pedestrian flows. The scenarios include corridors, doorways, intersections, and cluttered spaces populated with dynamic pedestrians. For each scenario, we independently vary passage width (or obstacle density) and pedestrian number, enabling evaluation of planners' capabilities in navigating narrow or cluttered spaces while maintaining appropriate social distance from both the target and surrounding pedestrians.