Learning Interaction-Aware Path Corrections from Unlabeled Trajectories: A LiDAR-Based Sim-to-Real GNN Framework for Construction Robots

Yilong Chen; Yong K. Cho

cs.ROOct 25, 2026

Learning Interaction-Aware Path Corrections from Unlabeled Trajectories: A LiDAR-Based Sim-to-Real GNN Framework for Construction Robots

Authors: Yilong Chen, Yong K. Cho

Organizations: School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA30332, United States

Abstract

Construction robots operating near workers require navigation that can anticipate human motion while remaining robust to dynamic and unstructured site conditions. However, existing social navigation methods depend on hand-crafted interaction rules, labeled demonstrations, or robot-specific control policies, limiting their scalability and transferability in construction environments. We propose a real-time, LiDAR-only framework that learns interaction-aware path corrections from unlabeled demonstrations. Constant-velocity extrapolation generates residual learning targets, while a GNN models relationships among the robot, pedestrians, and goal context. During deployment, an extended Kalman filter tracks LiDAR detections to construct dynamic graphs, and the GNN offsets planner-generated waypoints through a planner-agnostic interface. Within the bricklaying trial, the GNN reduced path irregularity by 39.2% and increased minimum TTC. Across Dynablox sequences, it increased minimum TTC by 38.3% while producing shorter, smoother paths than the baselines. Ablation studies supported a 1.0-s prediction interval and confirmed the benefits of target context, reflection augmentation, and feature normalization.

Figures & tables

Figure 1: Overview of the proposed interaction-aware path correction framework. During training, residual targets are automatically generated from unlabeled human demonstrations using constant-velocity extrapolation, and a GNN learns relationships among the agents after scenario-to-graph translation. During deployment, Dynablox detections from LiDAR point clouds are associated and tracked using EKFs to construct dynamic graphs online. The trained GNN then predicts spatial corrections to refine planner-generated local waypoints in real time.

Figure 2: Simulated Environment with HRI Patterns: (a) crossing paths, (b) merging, (c) overtaking, and (d) group interaction.

Figure 3: Scenario-to-graph translation.

Feature	dimension	details
Node	5	$[x,y,v_{x},v_{y},r]$
Edge	7	$[x,y,v_{x},v_{y},d,\sin\theta,\cos\theta]$

Table 1: Node and edge features

Figure 6: Bricklaying Experiment: (a) testbed layout and (b) quadruped robot with LiDAR sensor.

Figure 7: Dynablox dataset: (a) corridor, (b) hg, (c) ramp, and (d) stairs.

Dimension	Metric	Unit
Efficiency	path length	$meter$
Smoothness	path irregularity	$radian/meter$
Physical safety	closest pedestrian distance	$meter$
Social compliance	minimum time-to-collision	$second$

Table 2: Evaluation metrics

Measure	Traj.	Dataset
		corridor1	corridor2	hg1	hg2	ramp1	ramp2	stairs	Dynablox avg.	bricklaying
$\begin{array}[]{l}\text{Path}\\ \text{length}\\ (m)\downarrow\end{array}$	Origin	33.67	33.82	32.68	34.61	51.89	31.95	105.15	46.25 $\pm$ 24.89	58.96
	GNN	40.72	35.24	31.30	35.06	53.99	33.99	113.92	49.17 $\pm$ 27.33	63.25
	SFM	35.26	37.13	40.91	36.61	62.46	34.16	114.85	51.63 $\pm$ 27.35	63.97
	MLP	36.39	42.10	36.90	41.64	62.36	54.37	122.65	56.63 $\pm$ 28.36	82.65
$\begin{array}[]{l}\text{Path}\\ \text{irregularity}\\ (rad/m)\downarrow\end{array}$	Origin	0.26	0.29	0.61	0.47	0.22	0.24	1.79	0.55 $\pm$ 0.52	7.22

Table 3: Evaluation results on the Dynablox dataset and bricklaying trial. ”Dynablox avg.” reports the mean across seven sequences. Bold indicates the best result. Standard deviations are calculated within each sequence and across sequences for ”Dynablox avg.”.

Figure 8: GNN-learned HRI Patterns: (a) crossing paths, (b) merging, (c) overtaking, and (d) group interaction.

Measure	$\Delta t\quad(s)$	Dataset
		Dynablox avg.	bricklaying
$\begin{array}[]{l}\text{Path}\\ \text{length}\\ (m)\downarrow\end{array}$	0.5	48.33 $\pm$ 26.07	60.18
	1.0	49.17 $\pm$ 27.33	63.25
	1.5	51.10 $\pm$ 25.64	65.81
	2.0	58.98 $\pm$ 28.60	88.11
$\begin{array}[]{l}\text{Path}\\ \text{irregularity}\\ (rad/m)\downarrow\end{array}$	0.5	0.94 $\pm$ 0.35	5.05

Table 4: Results of ablation study on different prediction interval

\Delta t

. ”Dynablox avg.” reports the mean across seven sequences. Bold indicates the selected prediction interval. Standard deviations are calculated within each sequence and across sequences for ”Dynablox avg.”

Measure	Model variant	Dataset
		Dynablox avg.	bricklaying
$\begin{array}[]{l}\text{Path}\\ \text{length}\\ (m)\downarrow\end{array}$	full	49.17 $\pm$ 27.33	63.25
	w/o target node	47.50 $\pm$ 25.06	63.05
	w/o data aug.	48.99 $\pm$ 25.06	62.58
	w/o normalization	47.81 $\pm$ 25.27	61.95
$\begin{array}[]{l}\text{Path}\\ \text{irregularity}\\ (rad/m)\downarrow\end{array}$	full	1.39 $\pm$ 0.26	4.39

Table 5: Results of ablation study on different model variants. ”Dynablox avg.” reports the mean across seven sequences. Bold indicates the selected model variant. Standard deviations are calculated within each sequence and across sequences for ”Dynablox avg.”

Social Robot Navigation

Social robot navigation studies how robots move safely and comfortably around people while respecting human behavior, preferences, and shared-space norms.

See all 81 papers

Learning Social Robot Navigation By Sensing Human Legs
Jul 30, 2026Alberto Vaglio, Andrea Garulli, Antonio Giannitrapani +2RL for Robotics Robot Policy Learning
Department of Information Engineering and Mathematics, University of Siena, Siena, Italy · Uninettuno University, Rome, Italy
CORNAV: Construction-Aware Reasoning for Robot Navigation on Active Worksites
Oct 2, 2026Parastoo Ali Pour, Deepak Prakash Kumar, Tommy Zhou +2Robot Navigation Safe Robot Navigation
Department of Mechanical and Aerospace Engineering, University of California at Irvine, Irvine, CA 92697 USA · Department of Aerospace Engineering, Indian Institute of Technology Bombay, Mumbai 400076, India · Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697 USA
Towards Path-Creative Navigation: Robot Navigation through Embodied Interaction
Oct 8, 2026Haoyu Xi, Siwei Cheng, Xiangyuan Liu +2Robot Navigation Embodied Navigation
School of Computer Science, Shanghai Jiao Tong University, Shanghai, P. R. China · College of Information Science and Technology, Eastern Institute of Technology, Ningbo, P. R. China · School of Mechanical and Power Engineering, Zhengzhou university, Zhengzhou, P.R.China +1

Jul 30, 2026J/K move · Enter open · S save

Learning Social Robot Navigation By Sensing Human Legs

Alberto Vaglio, Andrea Garulli, Antonio Giannitrapani, Renato Quartullo+1

Department of Information Engineering and Mathematics, University of Siena, Siena, Italy · Uninettuno University, Rome, Italy

RL for Robotics Robot Policy Learning Robot Skill Learning Social Robot Navigation Robotic Perception

Open paper AskPDF arXiv

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.

Abstract

Social Robot Navigation

Explore similar work

Learning Social Robot Navigation By Sensing Human Legs

CORNAV: Construction-Aware Reasoning for Robot Navigation on Active Worksites

Towards Path-Creative Navigation: Robot Navigation through Embodied Interaction