Abstract
This paper presents and experimentally validates a concept of end-to-end imitation learning for autonomous systems by using a composite architecture of convolutional neural network (ConvNet) and Long Short Term Memory (LSTM) neural network. In particular, a spatio-temporal deep neural network is developed, which learns to imitate the policy used by a human supervisor to drive a car-like robot in a maze environment. The spatial and temporal components of the imitation model are learned by using deep convolutional network and recurrent neural network architectures, respectively. The imitation model learns the policy of a human supervisor as a function of laser light detection and ranging (LIDAR) data, which is then used in real time to drive a robot in an autonomous fashion in a laboratory setting. The performance of the proposed model for imitation learning is compared with that of several other state-of-the-art methods, reported in the machine learning literature, for spatial and temporal modeling. The learned policy is implemented on a robot using a Nvidia Jetson TX2 board which, in turn, is validated on test tracks. The proposed spatio-temporal model outperforms several other off-the-shelf machine learning techniques to learn the policy.
Highlights
With explosive developments in machine learning and control techniques during the past few decades, mobile robots are taking an increasingly important role in modern society
While Simultaneous Localization and Mapping (SLAM) is focused on geometry, deep learning has been proven to be capable of handling perception problems effectively by using a convolutional neural network (ConvNet) [3,4]
This paper presents a deep learning scheme by combining a convolutional neural network (ConvNet) [4] with a long short term memory (LSTM) [10] neural network for end-to-end learning, where the objective is to control and steer a mobile robot in a maze
Summary
With explosive developments in machine learning and control techniques during the past few decades, mobile robots are taking an increasingly important role in modern society. DQN and its various improvements [8] have shown at least human-level performance in these tasks To further develop these ideas and to use them in robotics and control applications, Zhu et al [9] implemented deep reinforcement learning (RL) to find a target in an indoor environment using only visual inputs. This paper presents a deep learning scheme by combining a convolutional neural network (ConvNet) [4] with a long short term memory (LSTM) [10] neural network for end-to-end learning, where the objective is to control and steer a mobile robot in a maze. The step is to incorporate memory into this system by using LSTM to increase the performance (e.g., accuracy) of the end-to-end network These two networks are combined to implement the trained model with Tensorflow [15] to perform real-time inference tasks and steer PSU BOT in a test environment.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
