Self-Driving or Biped Robot based on LSTM/RNN

Yesterday, our home-make robot can finally take some actions.
But We want it to learn how to take action with reinforcement learning and RNN/LSTM.
We’d like to build a self driving or walking robot.
Here are some questions that we’d like to know as soon as possible.

  1. Can DIGITs train the LSTM/RNN system?
  2. Is there any good example of reinforcement learning and RNN/LSTM that NVIDIA can share with us?

many Thanks.

Hello, any feedback?

  1. For model:
    DIGITs supports Caffe and Torch.
    If your model can be trained on Caffe or Torch, it should be able to train on DIGITs.

  2. For input/output:
    If you want to use DIGITs to convert input/output data into formal format, ex. lmdb/hdf5.
    Please make sure it is supported by DIGITs.

If your model can be trained on caffe but with its own input format.
You can generate lmdb file first, and then upload it to DIGITs for training.

Share an DIGITs example for text classification with Torch.
Although it doesn’t use recurrent-based model, it shows how to use DIGITs with the format other than image.