Machine Learning Projects

Robot Kinematics & Autonomous Car Racing

---

🦾 Project 1: Robot Kinematics Learning

Forward/Inverse Kinematics with ML
[Report] [Code]

Key Features

• Solved FK for 2DOF/3DOF/5DOF robots using 3 distinct approaches:
  • RBF Kernel + Linear Regression (2DOF)
  • Gradient Boosting (3DOF)
  • Deep Neural Network (5DOF)
• Jacobian matrix analysis (Frobenius norm = 2.995 vs analytical)
• Hyperparameter optimization with grid search

Performance Summary

DOF	Model	Position Error	Orientation Error
2	RBF + Linear Reg	1.78e⁻⁷ MSE	1.69e⁻⁶ MSE
3	Gradient Boosting	6.11e⁻⁵ MSE	0.29 MSE
5	Neural Network	0.0088 MAE	0.588 MAE

---

🏎️ Project 2: Autonomous Car Racing

Image-to-Action Classification
[Report] [Code]

Key Features

• 5-class CNN for racing control:
  • Input: 96x96x3 RGB images
  • Output: Steering/gas/brake actions
• Action space modeling:

  ACTIONS = {
      0: 'No action',
      1: 'Steer left',
      2: 'Steer right', 
      3: 'Accelerate',
      4: 'Brake'
  }

• Convolutional Neural Network (CNN) for classification
• Dual CNN Model to improve performance
• Data Augmentation to enhance model generalization
• Hyperparameter Search to optimize model performance
• Performance evaluation with multiple metrics (Accuracy, F1-Score, Precision, Recall)

Performance Summary - Single CNN

Dataset Configuration	Accuracy	F1-Score	Precision	Recall
Normal Data (Non-Aug)	46.95%	0.208	25.87%	29.27%
Normal Data (Aug)	48.51%	0.1	25.2%	20.19%
Marked Edges (Non-Aug)	49.12%	0.132	9.32%	24.12%
Marked Edges (Aug)	42.24%	0.13	21.57%	22.9%

Performance Summary - Dual CNN

Dataset Configuration	Accuracy	F1-Score	Precision	Recall
Normal Data (Non-Aug)	40.65%	0.383	46.88%	44.17%