Behavioral Cloning for Autonomous Navigation
This project leverages deep learning to enable the robot make decisions based on location or vision inputs
• Developed vision-based robot navigation systems using traditional machine learning and Convolutional Neural Networks (CNNs) in PyTorch.
• Implemented PCA for dimensionality reduction and StandardScaler for feature normalization to mitigate large feature space (12,288 features) and small sample size (400 samples) challenges.
• Developed a CNN model to process 64x64 RGB images, enabling the robot to recognize and navigate around obstacles.
• Optimized model training by adjusting hyperparameters, including learning rate, momentum, and batch size, resulting in faster convergence and improved performance.
• Conducted extensive testing to ensure the model's robustness and ability to generalize to new obstacle maps.