My team and I

Klay: AI-Powered Indoor Navigation

Leveraging the EfficientNet-b0 model and the Comma Body robotic platform, Klay introduces a pioneering solution for real-time, intelligent indoor navigation. This project morphs Comma Body into an astute navigation aide, capable of making informed directional decisions through various indoor environments by learning to move Left, Right, or Forward.

Data Collection:

Our team navigated Comma Body through various indoor scenarios, ensuring each captured image was labeled (L, R, F) based on the forced direction post-capture, creating a rich dataset that intricately intertwines diverse indoor features with optimal navigation directions.

Model Training:

Preprocessing: Rigorous image augmentations, e.g., perspective shifts, color adjustments, were applied for enhanced model generalization.
Adaptation: The EfficientNet-b0 model was tuned, modifying its output layer for our 3-class problem.
Training Strategy: Adopted a freeze-training approach, refining the final classifier layer and maintaining pre-trained weights.

Impact:

IndoorNavBody not only demonstrates the efficacy of melding machine learning with robotics for autonomous navigation but also unfolds new possibilities in robotics, like complex navigational behaviors and human-robot interaction, paving the way for future advancements in AI-driven robotics.

Thank you to the Comma.ai team and the open-source community for their invaluable contributions to this project.