🤖 SelfDriveSuite: Vehicle Control and Scene Understanding

Note: This project is a culmination of my work for the "ROB 535: Self-driving Cars: Perception to Control" course, conducted from September to December 2023. The course provided a solid foundation in autonomous driving technologies, from perception to control, and this repository represents the integration and application of those concepts.

🎯 Goal

This a comprehensive project focused on developing algorithms for self-driving cars. This repository encompasses several critical components of autonomous driving technology, including:

• Adaptive Cruise Control: Implementing an adaptive cruise controller to maintain safe following distances and speeds.
• Model Predictive Control (MPC): Designing both linear and non-linear MPC for tracking reference trajectories and optimizing vehicle paths.
• All-Weather Scene Understanding: Achieving robust real-time, object recognition, detection, and scene segmentation in challenging conditions, such as low visibility, fast-paced scenario, and/or adverse weather.

⚙️ Prerequisites

• Libraries/Frameworks:
  • Numpy
  • Matplotlib
  • Scipy
  • CVXPY (for Convex optimization)
  • CasADi
  • pytorch

🛠️ Test/Demo

• Adaptive Cruise Control
  • Go to the directory 'Vehicle Control\Adaptive Cruise Control', and launch the jupyter notebook
• Trajectory tracking using Linear MPC
  • Go to the directory 'Vehicle Control\Trajectory tracking', and launch the jupyter notebook
• Car Overtaking using Non-Linear MPC
  • Go to the directory 'Trajectory Optimization\CarOvertaking', and launch the jupyter notebook
• Drag Racing using Non-Linear MPC
  • Go to the directory 'Trajectory Optimization\DragRacing', and launch the jupyter notebook
• Image Classification
  • Go to the directory 'Image Classification', and launch the jupyter notebook
• Object Detection
  • Go to the directory 'Object Detection', and run 'inference.py' or 'inference_video.py' following README there.
• Scene Segmentation
  • Go to the directory 'Scene Segmentation', and launch the jupyter notebook

📊 Results

📈 Adaptive Cruise Control

📈 Trajectory tracking (using Linear MPC)

📈 Car Overtaking (using Non-Linear MPC)

📈 Drag Racing (using Non-Linear MPC)

• Case 1:

• Case 2:

• Case 3:

• Case 4:

• Case 5:

📈 Image Classification (Cars vs Person vs None)

• Challenging because low-resolution 32x32 blurry RGB images...still achieved 90% accuracy

📈 Object Detection (Traffic sign Detection)

• Achieved ~48 mAP with an average FPS of 43

📈 Scene segmentation (Comprising 14 classes from urban scenario + background class)

• Achieved 81.2 mIoU