Learning Perception and Control for Robot Intelligence

PhD Thesis Proposal Defence

Title: "Learning Perception and Control for Robot Intelligence"


Mr. Siyi LI


Autonomous robots that can assist humans in the daily unstructured world have 
been a long standing vision of robotics and artificial intelligence. The past 
few years have seen major advances in many perception and control tasks using 
deep learning and reinforcement learning methods. In this thesis proposal, we 
propose to explore learning-based solutions to robotic tasks.

Our first attempt is building a unified benchmark for visual object tracking on 
the unmanned aerial vehicle (UAV) platform. We manually build a drone tracking 
dataset, consisting of a variety of videos with high diversity captured by 
drone cameras. We devise new motion models for the drone tracking scenario by 
explicitly estimating the camera motion in the tracking phase.

Collecting real-world data with robotic systems is generally expensive due to 
the hardware cost and the manual labeling effort. However, deep learning and 
reinforcement learning methods require a data-hungry training paradigm. We 
propose to address this issue by learning in simulation and achieve successful 
transfer to real world. For robotic perception task, we investigate instance 
segmentation for robot manipulation. We develop an automated rendering pipeline 
to generate photorealistic synthetic images with pixel-level labels. The 
synthetic dataset is then used to train an objectness deep neural network model 
which can successfully generalize to real-world manipulation scenarios. For 
robotic control task, we focus on the problem of learning UAV control for 
actively tracking a moving target. We propose a hierarchical approach that 
combines model-free reinforcement learning methods with conventional feedback 
controllers to enable efficient and safe exploration in the training phase. We 
show that this hierarchical control scheme can learn a target following policy 
in a simulator efficiently and the learned behavior can be successfully 
transferred to real-world quadrotor control.

Date:			Wednesday, 12 September 2018

Time:                  	3:30pm - 5:30pm

Venue:                  Room 4475
                         (lifts 25/26)

Committee Members:	Prof. Dit-Yan Yeung (Supervisor)
 			Dr. Yangqiu Song (Chairperson)
 			Dr. Raymond Wong
 			Prof. Nevin Zhang

**** ALL are Welcome ****