McMaster University
Learning Social Robot Navigation From Real-World Demonstrations With a Multi-Model Evaluation of Social Compliance
Abstract
dc:description.abstractAutonomous mobile robot navigation in human-populated environments requires more than simple obstacle avoidance; it demands adherence to uid social norms. This thesis investigates the e cacy of Learning from Demonstration (LfD) for developing socially compliant navigation policies. Utilizing a Turtlebot4 platform, a real-world dataset was collected on the McMaster University campus to train three distinct imitation learning architectures: Behavioral Cloning (BC), Generative Adversarial Imitation Learning (GAIL), and Di usion Policies. The models were evaluated using a multi-modal approach involving simulation metrics, video-based behavioral analysis, and human-robot interaction (HRI) surveys. Quantitative results indicated that while BC achieved high positional accuracy, it lacked robustness under distribution shifts. The Augmented Di usion Policy emerged as the superior architecture, maintaining a 91.7% success rate while achieving the lowest "Social Work" score, signifying minimal disruption to human tra c. However, HRI survey data revealed a "safety-perception paradox," where the robot's prioritization of social distance led to increased wall collisions, negatively impacting user perception of path consistency. This research concludes that social compliance is a multifaceted optimization problem requiring a balance between human-awareness, spatial safety, and mechanical predictability.
Degree
thesis:*- Department dc:contributor.department
- Computing and Software
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Taherifard, Fatemeh
- Advisors dc:contributor.advisor
-
- Geiskkovitch, Denise
- Stoyanov, Todor
Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.71548/21456