search

uzh-rpg / agile_autonomy

Repository Containing the Code associated with the Paper: "Learning High-Speed Flight in the Wild"
GNU General Public License v3.0
588 stars 164 forks source link

train neural network #91

Open w2385680516 opened 8 months ago

w2385680516 commented 8 months ago

Hello author, what trick did you use when training the neural network? In the forest scene, I used the train.py file directly, but the trained model could never achieve the same effect as your model. can you help me? Thanks.

antonilo commented 6 months ago

We trained on multiple scenes and a large number of environments. The quickest way to do this is to download the data from the associated link and train on our previously collected dataset. After a finetuning round at the speed you need to go, the network would reach approximately (accounting for the randomness in the simulator) the same performance of our checkpoint.