Smartphone-distracted pedestrians deviate from social norms unpredictably and cannot be relied upon to yield to approaching robots -- a safety challenge that existing navigation methods are poorly equipped to handle, since they treat all pedestrians uniformly regardless of attention state. We propose an attention-aware navigation framework that classifies each pedestrian as attentive or distracted in real-time using 3D skeletal features and a Random Forest classifier, then dynamically widens cost-map safety margins around distracted individuals to drive proactive avoidance. A within-subjects user study with 12 participants across 252 trials demonstrates significant improvements in perceived smoothness (p = 0.003) and predictability (p = 0.034) over both a LiDAR-only baseline and a trajectory-prediction-only method, without any increase in travel time. Notably, trajectory prediction alone does not improve subjective navigation quality unless coupled with attention-state-aware cost adjustment -- confirming that appropriately responding to who is in front of the robot can matter more than trajectory smoothness alone.
Figures & tables
Fig. 1 : An overview of the proposed attention-aware cost-map tuning. A robot secures a standard safety margin for an attentive pedestrian (right) and a larger margin for a smartphone-distracted pedestrian (left).
Fig. 2 : Framework of the proposed system. The system consists of three main modules: (1) pedestrian trajectory prediction based on historical position data, (2) Attention Classification using 3D skeletal keypoint extraction (33 points) from RGB-D camera images, (3) adaptive CostMap Generation using predicted trajectories based on attention state classification results.
Fig. 3 : Experimental setup in an indoor intersection (main corridor: 3.0 m × 10.0 m/8.0 m; cross corridor: 2.15 m × 7.2 m), 12 participants walk toward the robot.
Predicted
Actual
Attentive
Distracted
Attentive
35
2
Distracted
1
23
TABLE I : Confusion matrix (left) and classification performance metrics (right) using Random Forest
Fig. 4 : Feature importance of the Random Forest model. Feature importance scores indicating that wrist and shoulder positions are dominant factors.
Metric
A
B
C (Ours)
Jerk RMS (m/s 3 )
37.45 ± 17.46
11.01 ± 8.62
11.42 ± 9.19
Min separation (m)
1.74 ± 0.68
0.94 ± 0.28
1.01 ± 0.31
Max lat. dev. (m)
0.30 ± 0.21
0.31 ± 0.19
0.36 ± 0.23
Travel time (s)
16.8 ± 2.1
17.0 ± 1.9
16.9 ± 2.0
∗ Method A’s 39% detection rate is inflated; see Section V-B.
TABLE II : Quantitative evaluation results across three methods with mean ± SD.
Item
A
B
C (Ours)
Q1 Safety
4.60
4.69
5.07
Q2 Comfort
4.24
4.42
4.82
Q3 Smoothness
3.96
4.14
4.89
Q4 Naturalness
4.13
4.22
4.67
Q5 Predictability
3.69
4.01
4.48
Q6 Trust
4.49
4.64
4.96
TABLE III : Subjective evaluation results (Q1–Q7). Mean scores are shown for each method.
Safe robot navigation develops methods for reaching goals while avoiding collisions and managing uncertainty, dynamic obstacles, and environmental hazards.
Robots navigating among pedestrians typically sense their surroundings with a 2D LiDAR mounted close to the ground. At that height, the sensor mostly sees moving legs rather than whole people, yet most learning-based navigation methods still treat pedestrians as simple shapes like circles. This paper addresses that gap with CALF (Convolutional Attention for Leg Features), an end-to-end neural architecture that combines convolutional layers, attention, and MLP to interpret leg motion directly from LiDAR scans and produce safe navigation commands. The CALF policy is trained using deep reinforcement learning algorithms within LegNav, a custom lightweight 2D simulator that combines 2D LiDAR ray tracing with a novel pedestrian gait model. The resulting policy is compared against classical and learning-based baselines in terms of navigation performance and social compliance. The approach is validated through real-world experiments via zero-shot deployment on a TurtleBot 4, yielding smooth and socially compliant trajectories. Written in JAX, the LegNav simulator enables the training of a deployment-ready CALF policy in under an hour on a single consumer GPU.
Alberto Vaglio, Andrea Garulli, Antonio Giannitrapani +2
Department of Information Engineering and Mathematics, University of Siena, Siena, Italy · Uninettuno University, Rome, Italy
Legibility in human-robot interaction (HRI) is often described as the property of robot motion that enables an observer to confidently and quickly infer the robot's intent. Mature frameworks exist for generating legible motion in front of static observers, typically by making the robot's destination apparent as early as possible. Social robot navigation in public spaces, however, presents a different challenge: pedestrians are themselves in motion, often with divided attention, and must coordinate with the robot in real time. In such encounters, where the robot is ultimately headed may matter less to a pedestrian than how it intends to pass them. This raises two questions: which aspect of intent should legible motion convey, and does legibility remain beneficial when pedestrians are not fully attentive to the robot? We address these questions in the ubiquitous and demanding scenario of hallway navigation, through two controlled user studies comparing multiple legibility formulations implemented within a shared model predictive control framework. Study 1 (N = 45) compares destination-based and passing-side intent representations, showing that the latter, particularly when adaptively updated, allows for smoother human motion and is perceived as more competent and less mentally and physically demanding than the former. Study 2 (N = 45) examines the effect of pedestrian attention, demonstrating that legible motion supports smooth human motion even under distraction, although this benefit is not consistently reflected in subjective ratings. Together, these findings suggest that effective legible motion in social robot navigation benefits from interaction-level intent representations that support coordination, with some effects persisting even when human attention is divided.
Pranav Goyal, Andrew Stratton, Christoforos Mavrogiannis
Many pedestrian trajectory prediction algorithms have been proposed to improve the safety of navigation for mobile robots working in human-robot coexistence environments. Some pedestrian trajectory prediction algorithms extract information about obstacles near pedestrians from top-down view images to improve the accuracy of trajectory prediction. However, mobile robots typically create local occupancy maps using LiDAR, rather than top-down view images. Meanwhile, the vision sensors on board robots provide egocentric view images, which contain fine-grained behavioral information about the pedestrians near the robot. To better use the information collected by LiDAR and on-board vision sensors, we propose MamMA, a Mamba-based pedestrian trajectory prediction algorithm considering occupancy maps and pedestrian awareness states. MamMA divides the occupancy map by patches and extracts obstacle features from each patch to create map features. Pedestrian awareness states are divided and considered, as some studies show that awareness states affect the perception and speed of pedestrians. Furthermore, a Mamba-based model is proposed to predict the future trajectories of pedestrians based on different types of features. Experiments on the STCrowd, SiT, JRDB, ETH, and UCY datasets show that MamMA achieves better average displacement error and final displacement error than the state-of-the-art algorithms.
Juncen Long, Xiaofeng Jin, Gianluca Bardaro +2
Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milan, Italy