Development blog for TORCS project.
by Oliver
This week was the week we presented a demonstration of our AI racer to our client, John McNamara, so we had no facilitator meeting. Our Presentation was late Wednesday afternoon, so we planned an all-day team meeting to complete, perfect and practice our demonstration to prepare ourselves as best we could. Since our facilitator meeting the previous week, we had began work on our 15 minute presentation, inclunding a live demonstration of our 1:32:30 lap time, and a detailed description of our progress through the project requirements.
During this week, with the presentation looming ahead of us, we continued training our models as much as we could, before capturing a final lap time. In our previous facilitator meeting, we were at a lap time of 1:39:13. However, while investigating how to change to the IBM livery for the video demonstration, we discovered that Kerem’s installation of The Open Race Car Simulator (TORCS) did not natively use the F1 car. This meant the model had been being trained on a slower car. After looking at installation and folders, we managed to locate the file that automatically loaded the car, and changed it to the F1 car. Immediately, the lap time increased, and by training the model with this car until Wednesday, we improved our lap time by almost 7 seconds, to 1:32:30.
At 15:30 on Wednesday the 6th of May, we stepped into the meeting room with John. After some friendly introductions, the start of our presentation was, unsurprisingly, hindered by some technical difficulties. Though, once there were resolved, we were swiftly underway. We introduced ourselves and described our roles within the team, then spoke to John about the state of our project in relation to the system requirements - he was pleased to learn we were quite ahead! Following that, was a detailed description of our models, how they work, and their progress within training. After explaining the remaining features that we need to implement, and playing the video demonstration of our lap time, John provided us with some excellent feedback. He advised us to enter the IBM National Competition for this AI Racer Project, as we would have a good chance of placing well, and that of the verified lap times he has seen, we are very competitive. Given we still have 2 remaining weeks until the project deadline, we assured him that further training would even further reduce our lap time!
Now that we are in the final stages of development, we will need to focus on implementing the last few system requirements, and improving the ease of use of the project, e.g. installation. These are the remaining tasks we need to complete: