How hard is it to cross the room? -- Training (Recurrent) Neural Networks to steer a UAV

02/24/2017
by   Klaas Kelchtermans, et al.
0

This work explores the feasibility of steering a drone with a (recurrent) neural network, based on input from a forward looking camera, in the context of a high-level navigation task. We set up a generic framework for training a network to perform navigation tasks based on imitation learning. It can be applied to both aerial and land vehicles. As a proof of concept we apply it to a UAV (Unmanned Aerial Vehicle) in a simulated environment, learning to cross a room containing a number of obstacles. So far only feedforward neural networks (FNNs) have been used to train UAV control. To cope with more complex tasks, we propose the use of recurrent neural networks (RNN) instead and successfully train an LSTM (Long-Short Term Memory) network for controlling UAVs. Vision based control is a sequential prediction problem, known for its highly correlated input data. The correlation makes training a network hard, especially an RNN. To overcome this issue, we investigate an alternative sampling method during training, namely window-wise truncated backpropagation through time (WW-TBPTT). Further, end-to-end training requires a lot of data which often is not available. Therefore, we compare the performance of retraining only the Fully Connected (FC) and LSTM control layers with networks which are trained end-to-end. Performing the relatively simple task of crossing a room already reveals important guidelines and good practices for training neural control networks. Different visualizations help to explain the behavior learned.

READ FULL TEXT

page 2

page 6

page 15

page 17

research
09/23/2022

Image Classification using Sequence of Pixels

This study compares sequential image classification methods based on rec...
research
08/09/2017

Tikhonov Regularization for Long Short-Term Memory Networks

It is a well-known fact that adding noise to the input data often improv...
research
09/24/2018

Vision-based Control of a Quadrotor in User Proximity: Mediated vs End-to-End Learning Approaches

We consider the task of controlling a quadrotor to hover in front of a f...
research
07/02/2019

Kite: Automatic speech recognition for unmanned aerial vehicles

This paper addresses the problem of building a speech recognition system...
research
05/15/2018

Characterizing the Temporal Dynamics of Information in Visually Guided Predictive Control Using LSTM Recurrent Neural Networks

Theories for visually guided action account for online control in the pr...
research
03/03/2018

Teaching UAVs to Race With Observational Imitation Learning

Recent work has tackled the problem of autonomous navigation by imitatin...
research
08/13/2020

Effect of Architectures and Training Methods on the Performance of Learned Video Frame Prediction

We analyze the performance of feedforward vs. recurrent neural network (...

Please sign up or login with your details

Forgot password? Click here to reset