This paper provides an overview of existing solutions for autonomous driving within the CARLA simulator, which are based on end-to-end mode of operation. A new system was developed which is also based on the end-to-end mode of operation using the CARLA simulator as the simulation environment. The developed system uses a combination of several inputs to successfully predict the car's steering, i.e. the gas, brake and steering angle values are predicted. Inputs are RGB and depth camera images and information about current vehicle speed and turn commands. The aim of the work is to design, train and evaluate a model with the goal of autonomous driving. The model was trained on data collected in the simulation environment CARLA simulator, and the evaluation was performed on three maps. Two of the three evaluation folders are not part of the overall database, while one folder is part of the database used for training. The neural network consists of multiple convolutional layers that process RGB and depth camera visual data while vehicle speed and turn commands are fed into fully connected layers. The evaluation results showed that the model copes well in normal driving conditions, i.e. in simpler traffic situations, while in more complex traffic situations mistakes are possible when driving the car. Also, the number of errors made by the model while driving on a known map is significantly lower than the number of errors on unknown maps, which indicates the need to expand the database with different driving conditions. This paper shows the possibilities of applying convolutional neural networks in the field of autonomous driving using a combination of sensor data. The model that uses a combination of sensor data showed better results during the evaluation compared to the model that uses only one sensor, the RGB camera.