A paper co-authored by Ph.D. candidates Seokjoon Kwon and Hee-Deok Jang, along with Professor Dong Eui Chang from our school’s Professor Dong Eui Chang’s Lab, has won the Excellence Paper Award at the Korea Artificial Intelligence Association (KAIA) Fall Conference.
The award-winning paper, titled “A Framework for Object Navigation Using Hierarchical Scene Representations,” proposes the HSG-ON (Hierarchical Scene Graph-based Object Navigation) framework to solve the Zero-Shot Object Goal Navigation task which is the challenge of an AI robot finding an object it has never encountered in an unfamiliar environment.
The framework constructs a Hierarchical Scene Graph (HSG) that represents the environment in a ‘room-workspace-object’ hierarchy. Based on HSG, the robot is enabled to strategically search the ‘workspace’ most likely to contain the target object. By narrowing the search space in a human-like manner, HSG-ON demonstrates superior performance in terms of success rate and efficiency compared to conventional methods, offering a powerful and efficient solution for the Zero-Shot Object Goal Navigation task.
