Deep Learning in Robotics: A Review of Recent Research

Four broad categories of Neural Networks

A) Feed-Forward network (Function approximator)

B) Auto-Encoder (Encoder/Decoder)

C) Recurrent Network

D) Q-Learning network

Seven challenges they see in robotics that could be addressed by DL

1. Learning complex, high-dimensional, and novel dynamics.
2. Learning control policies in dynamic environments.
3. Advanced manipulation.
4. Advanced object recognition.
5. Interpreting and anticipating human actions
6. Sensor fusion & dimensionality reduction.
7. High-level task planning.

From the section “Practical recommendations for working with Structure D”

Another important technique is to train in simulation before attempting to train with an actual robot. This reduces wear on physical equipment, as well reduces training time. Even if only a crude simulation is available, a model that has been pre-trained on a similar challenge will converge much more quickly to fit the real challenge than one that was trained from scratch.

They also suggest using more intelligent exploration strategies than epsilon-greedy.