reinforcement learning approach:

1. Based on CNN : different behaviour for different detection;
2. Based on LSTM/RNN: improve navigation in labyrinth-like scenario;
3. Know target research in a labyrinth-like scenario;
4. Run-time training;