The ability for robots to successfully navigate indoors is a critical step to bring the benefit of robotics to the general population. We use deep imitation learning on the THOR dataset to demonstrate the potential for deep imitation learning, in which an expert provides the learner the best routes and expects the learner to extrapolate to new situations to teach robots how to navigate indoors. Using the DAGGER algorithm, we train a neural network on one indoor scene and validate its effectiveness in a different indoor scene. We show that the DAGGER algorithm is able to substantially improve the average trajectory length compared to other potential navigation algorithms when generalizing to navigation in previously unexplored indoor scenes.
News Source |
Deep DAgger Imitation Learning for Indoor Scene Navigation
RELATED NEWS
PUBLICATIONS AND PRESS
News Source
Deep Learning Specialization
Become a Machine Learning expert. Master the fundamentals of deep learning and break into AI....
Read more...News Source
Natural Language Processing Specialization
Break into the NLP space. Master cutting-edge NLP techniques through four hands-on courses!...
Read more...News Source
Using Spatio-Temporal Data To Create A Shot Probability Model
Basketball is an invasion sport, which means that players move freely during every second of the gam...
Read more...