Page 90 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Electronic & Computer Engineering

It is desirable to have a single-wavelength narrowband optical source for high-data-rate optical communications, but this
kind of light source are very expensive. A general laser pointer can only emit multiple wavelengths within a spectral
band typically. It is hoped to develop that single-wavelength narrowband optical source by engineering a multiple-
wavelength semiconductor laser chip from a laser pointer and combine the light into an external cavity through this
project. We try different combinations of cavity thickness, tilted angles, feedback strength and voltage and want to find
a combination that giving out a spectrum close to a single-wavelength narrowband optical source. It is aimed to give a
stable single-wavelength laser source with low cost by the resulting system.

Autonomous Driving Using Deep Learning

Supervisor: SHEN Shaojie / ECE
Student: DEL MUNDO Gian Miguel Sero / COMP Course: UROP1100, Spring

Autonomous driving has recently become a hot topic in computer science, but state-of-the-art approaches have not
been able to account for non-deterministic scenarios, semantic information, interactions with other agents such as
pedestrians, or long-term possibilities. The project explores a proposal by Lee et al. called the Deep Stochastic IOC RNN
Encoderdecoder network (DESIRE) to overcome the aforementioned limitations. It consists of two modules, namely the
Sample Generation Module (SGM) and the Ranking & Refinement Module (RRM). The SGM produces a diverse set of
future trajectory predictions based on the past trajectory of the car, and the RRM scores these predictions based on
semantic information retrieved by a convolution neural network (CNN) as well as interactions with other agents

Supervisor: SHEN Shaojie / ECE
Student: YANG Handi / CPEG Course: UROP1100, Spring

To derive the optimal strategic long-term future prediction in driving behavior, a rich pool of potential samples is the
foundation for later ranking and refinement. Based on the paper, “DESIRE: Distant Future Prediction in Dynamic Scenes
with Interacting Agents”, we reproduce the Diverse Sample Generation Module which accounts for past motion histories
on the task of autonomous driving. To tackle the uncertain nature of prediction, a Conditional Variational Auto-encoder
(CVAE) is adopted to produce a set of hypothetic paths with introduction of a latent variable. Results of plausible
trajectories are compared with the ground truth and baseline methods to evaluate the performance.

Robust Monocular Visual-inertial Localization and Mapping

Supervisor: SHEN Shaojie / ECE
Student: ZHANG Yu / CPEG Course: UROP1100, Spring
UROP2100, Summer

This report briefly summarizes the work done on the UROP project entitled “Robust monocular visual-inertial
localization and mapping” under Professor Shen Shaojie in Summer 2018. Most of the work was done on the subproject
named mockaswarm which was an all-in-one system for simulation, visualization and control of a swarm of quadrotor
drones. In order to simulate the missions inside the system, a map generation module was improved which includes
optimizing the map by removing excess points in order to reduce map data size, generation of a random 2D maze with
variable size and generation of a random 3D maze with variable size, cell number and connectivity.

80