Path Generator with Unpaired Samples Employing Generative Adversarial Networks.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Robotics

Background:

• Interactive technologies like augmented reality (AR) demand significant computational power and specialized sensors for real-time immersive experiences, leading to high implementation costs.
• Machine learning offers cost-reduction potential but faces challenges due to the complexity of creating comprehensive datasets for environmental perception.

Purpose of the Study:

• To propose an alternative strategy for real-time path generation on embedded devices using limited, unpaired environmental data.
• To enable augmented reality experiences that can navigate physical spaces by avoiding virtual elements, even with imperfect environmental knowledge.

Main Methods:

• Utilized unpaired samples from known and unknown surroundings to generate navigation paths.
• Developed an architecture for path creation based on imperfect environmental knowledge.
• Integrated the generated path into an augmented reality experience for user testing and performance evaluation.

Main Results:

• Successfully approximated a navigation path using an unpaired dataset, demonstrating feasibility with limited information.
• The proposed strategy allows for real-time path generation on devices like smartphones.
• User testing validated the performance of the augmented reality experience with the generated path.

Conclusions:

• The primary contribution is the approximation of a path using unpaired data, offering a cost-effective solution for real-time navigation in augmented reality.
• This method addresses the limitations of high costs and complex dataset creation in current interactive technologies.
• The research paves the way for more accessible and efficient augmented reality applications on mobile devices.