ChauffeurNet : Learning to Drive by Imitating the Best and Synthesizing the Worst. Reproduction the result according to this paper[https://arxiv.org/pdf/1812.03079.pdf]. I just implement it on the basis of my comprehension,because the paper didn't introduce the neural network in every detail. The model is implemented by Keras with Tensorflow backend.
1.Model and train and prediction with mocked data.[done] 2.Data pipeline for real data. 3.Train it in real world data. 4.Other approachs in paper. 5.Test it in simulation. I want the model can be used in different simulation environment. Welcome other contributors to integrate different open source or private simulators. I will combine my company's simulator and some simple simulators first. 6.Test it in Real world on china's urban road.
1.use conv layers like U-Net(Conv+Upsampling/Deconv) [done] 2.Conv + Full Connect like artari-net 3.Fully Conv 4.Fully Conv + GRU
https://github.com/Iftimie/ChauffeurNet
anaconda3, python 3.6, keras 2.2.4, tensorflow 1.12.0
python chaffeur_net/chaffeur_net_main.py ###rnn_cell is reserverd. we should not use it now!