AI Drive

A self-driving car that drives around aimlessly. It has been trained on 16 labeled examples in TensorFlow with Keras to navigate it's environment while minimizing its chances of getting stuck. 16 weights and 4 biases were then ported over to this TerraTech circuit, adding up to a total of 20 parameters. A single layer of 4 ReLU neurons are used in a feed-forward architecture in this neural network, with any output at or over 0. 5 rounded up to 1, otherwise 0. There is one problem however where a hard rule is applied instead of the neural network's judgement. When all sensors but the front are activated, the neural network fails to act and instead only outputs zeros. This has been fixed for the time being by using a small circuit that detects the particular sensor activation combination and moves the tech forward. Otherwise, all other decisions are made entirely by the AI. If you wish to add the self-driving AI to your techs, press X on the tech and remove the car and snapshot the neural network part of the tech. Place a proximity sensor at the front, back, left, and right of the tech you wish to add the AI to and make sure they are just big enough to cover the side of the tech they are on completely. Then add a transmitter to each proximity sensors' output and use channel 1 for front, channel 2 for back, channel 3 for left, and channel 4 for right. After that, add a tech controller block and place a receiver tuned into channel 10 on forward, 20 on backward, 30 for left, and 40 for right. Known issue: The neural network also appears to be unresponsive when both side sensors are activated but the front and back sensors are not. This can be solved by connecting channels 1 and 2 receivers to an AND gate, connecting channels 3 and 4 to two NOT gates, connecting the output of those two NOT gates to an AND gate, then connecting the two AND gates to another AND gate connected to a channel 10 transmitter. I figure I might have to reupload this circuit just to solve this issue