Diskussion:JetRacer: Autonomous Driving, Obstacle Detection and Avoidance using AI with MATLAB

Task: Weekly Progress Report

Write down, what you did and what you want to do next week. Diskuss this in the weekly meeting with Prof. Schneider

#	date	plan for the next week	work progress
1	20.01.2025	Dokument you progress in the article. Plan you task as a gantt-chart and discuss it with Prof. Schneider time report in Sciebo next steps:	The steering angle ranges from -20°..20° (1..-1). The normalized steering range 1..-1 has the wrong sign. A left curve has an steering angle of 2° (-0.1). A richt curve has an steering angle of -2° (0.1). Programm the controller that you cann add small steering angles by pressing a button. One Button für straight, left -0.05, right +0.05.
2	22.01.2025	Dokument you progress in the article! Record the video while driving in best quality and process it with MATLAB®. Rectify the video with MATLAB® camera calibration app. Convert the video to different formats and teach the CNN. What works best? Optimize the steering - smooth steering approach 1: press button approach 2: use old RC-Controller potentiometer	Meeting with Mr. Christopher Beck The approach with steering angle and video is promissing. Research different input data. How can the algorithm learn best? RGB, Gray, BW, Canny-Edge, ROI, Birdeyeview Teach the CNN with more data. Find out the best modell with [1]
3	23.01.2025	Dokument you progress in the article! Calibrate the camera over the whole field of view. Test the calibration with the checkmate on the ground. Measure height and pitch angle of the camera. Use Messdaten2MAT.m to convert the steering_values.csv to a mat-file. Display the video, the input steering angle and the CNN steering angle with zeigeMessdaten.m. A red circle or vertical line visualizes the time frame in the data.	Succesffully recorded the driving with the best driving quality with approach .1(driving by incrementation) calibrated the camera with the camera calibration App. Used checkmate on the ground to test the camera calibration parametters aswell as using grides(vertical lines + horizontal lines) Teach the CNN with more data.
4	28.01.2025	Dokument you progress in the article! Camera height: ~14cm, pith angle: ~20° Read the exact values from the calibration with a flat checkmate image. Record the video while driving in best quality and process it with MATLAB®. Save it in SVN as YYMMDD_Circuit_rgb.mp3 or avi. Transform it with MATLAB® to _rgb_rect.mp4 _gray.mp4 _gray_rect.mp4 _bw.mp4 _bw_rect.mp4 _CannyEdge_rect.mp4 _BirdEye_rect.mp4 Tutorial: SoSe24_-_Praktikum_Systementwurf_-_Inverse_Perspektiventransformation_(IPT) Save the data to SVN. Teach the AI with these videos seperately. What Input gives the best results? What are the best AI parameters? Is the steering angle input smooth and not bouncing?	Succesfully saves video on Svn as mentioned Transformed the video from avi to (_rgb_rect.mp4. _gray.mp4, _gray_rect.mp4, _bw.mp4, _bw_rect.mp4, _CannyEdge_rect.mp4 and _BirdEye_rect.mp4) Saved data to SVN Configured different parametters to teach the AI with these videos seperately to determine the best resault
5	03.02.2025	Dokument you progress in the article! Wrote to professor to reschedule our weekly meeting. Still working on AI training task with different parameters Follow the tasks above: First record the video in best quality @30fps (e.g. 3280x2464 pixel) and then convert the video to AI Video formats. Steer in 0,05° steps.	Succesfully transformed the video from avi uncompressed to (_rgb_rect.mp4. _gray.mp4, _gray_rect.mp4, _bw.mp4, _bw_rect.mp4, _CannyEdge_rect.mp4 and _BirdEye_rect.mp4) Saved data to SVN Configured different parametters to teach the AI with these videos seperately to determine the best resault
6	10.02.2025	Teach the AI with these videos seperately. What Input gives the best results? What are the best AI parameters? Is the steering angle input smooth and not bouncing?	Teaching the model with different parameters CNN Finetuning Pretrained model(Resnet-18) RNN LSTM(Long Short-Term Memory) Reinforcement Learning Transfer Learning
7	17.02.2025	Object detection and Object tracking algorithm Implement a maneuvring movement of the jetracer to switch between lanes	Please document your progress at least once a week.
8	25.02.2025	Student is stuck with the AI-teaching - Task: make appointment with Mr. Beck, present the results and ask for a hint. Student wants to implement a classic image processing based Lane Following algorithm. We discussed the necessary steps and Prof. Schneider offered to provide a framework. Student has to send a SVN-link of the input video and the camera calibration file (*.mat) to Prof. Schneider.	Please document your progress at least once a week.