Towards Behavioural Cloning for Autonomous Driving

Kumaar, Saumya and Navaneethkrishnan, B and Hegde, Sinchana and Raja, Pragadeesh and Vishwanath, Ravi M (2019) Towards Behavioural Cloning for Autonomous Driving. In: IEEE INTERNATIONAL CONFERENCE ON ROBOTIC COMPUTING (IRC 2019), 25-27 Feb. 2019, Naples, Italy, pp. 560-567.

PDF IRC_2018.pdf - Published Version Restricted to Registered users only Download (793kB) | Request a copy

Abstract

This paper proposes an off-policy imitation learning methodology for autonomous driving using a doubly-deep recurrent convolutional architecture that learns compositional representations in both space and time domains. The architecture has been referred to as NAVNet (Navigation Network) and is end-to-end trainable. The recurrent long-term models are directly connected with the visual convolutional models. The models can be trained together to learn both temporal dynamics as well as the convolutional perceptual representations. The approach is non-data driven in nature and the system learns a regression-based mapping function between input images and steering angle. Results presented in this research indicate distinct advantages of the proposed LRCN model over the state-of-the-art deep learning techniques for autonomous navigation.

Item Type:	Conference Proceedings
Series.:	IEEE INTERNATIONAL CONFERENCE ON ROBOTIC COMPUTING
Publisher:	IEEE
Additional Information:	Copyright for this article belongs to IEEE
Keywords:	Autonomous Driving, Unmanned Aerial Vehicles, Convolutional Neural Networks, LSTM, Behavioral Cloning
Department/Centre:	Division of Mechanical Sciences > Aerospace Engineering(Formerly Aeronautical Engineering) Others
Date Deposited:	23 May 2019 11:56
Last Modified:	23 May 2019 11:56
URI:	http://eprints.iisc.ac.in/id/eprint/62747

Actions (login required)

View Item