Problem Statement (MPC): In this project, an F1 vehicle is subjected to a Miami race-track consisting of randomly generated obstacles in its course. The defined vehicle is bounded by speed and steering angle constraints, with a goal to complete the track in minimum time and avoid all the random obstacles. The vehicle can detect an obstacle within a range of 250 m.

Problem Statement (Image Classification): From the given dataset of 10,000 images comprising of 22 classes, develop an image classification model with an accuracy of more than 45%.

MODEL PREDICTIVE CONTROL:

• In the first half of the project, a PI controller is used in synthesizing a trajectory which allows the vehicle to race through the given state and vehicular constraints.
• A novel Model Predictive Controller scripted in MATLAB is deployed on the said trajectory.
• The controller senses the obstacles within a 250m range, and enforces a lane-changing maneuver on the given two lane-track. Challenges and Drawbacks:
• In instances where the obstacles are generated near sharp turns or when two obstacles are generated very near to each other in alternate lanes, the controller either collides into an obstacle or leave the track.
• This is because of the fact that the vehicle is majorly driven through steering with constant speed. Thus above instances are extremely difficult to pass through.
• This serves as the scope for the research project.

IMAGE CLASSIFICATION:

• For the purpose of simplicity, a SSD300 with VGG16 is selected as the backbone of the model.
• With a filter size of 5x5 and padding of 2, the model is loaded with training image set of 7,000 images.
• On the calculated weights, the remaining 3,000 images are tested and evaluated to achieve an accuracy of 51%.

Link to GitHub Repository