About Me
Nice to meet you! I am Yuanchu Liang based in Australia, Canberra. I am doing a computer science PhD, supervised by Prof. Hanna Kurniawati, at the Robust Decision Making and Learning Lab, The Australian National Univeristy (ANU). Before this, I finished my high school in Adelaide, and graduated with a double degree in Engineering and Science from the ANU.
My research direction focuses on building robust general purpose sequential decision making systems with application in robotics. It is my hope to see robots becoming an essential ingredient in our daily life in the future, just like smart phones today. To achieve this, I want to extent robots’ capabilities in handling uncertainties from the real world and efficiently learning useful patterns from the environment.
In particular, I am interested in the Partially Observable Markov Decision Process (POMDP) and designing scalable POMDP solution methods for Robotics motion planning under uncertainties. On the learning aspects, I look into sample efficient reinforcement learning algorithms, transfer learnings and generalisations under uncertainties.
Outside of work, I love meeting new people, sharing new ideas, reading books and climbing rocks, the latter has has became my way of meditation :)
Education
The Australian National University
Double Degrees in Bachelor of Engineering (First Class Honours) and Bachelor of Science.
2019 - 2023
The flexibility offered by ANU allows me to do a double degree in Enginnering and Science.
I specialised in mechatronics engineering and explored areas like control theory, system dynamics, embedded systems and robotics.
Under my science degree, I studied computer science with a focus on AI and ML, and mathematics including analysis, algebra and computational maths.
Publications
This is my first PhD project. I am lucky enough to collabote with Lydia Kavraki’s lab to propose a novel algorithm that leverage fast motion planning library (VAMP) and the reference-based POMDP to handle online POMDP planning with long horizons. Empirically, we test on many robotic problems and our method outperforms the state-of-the-art methods by multiple factors.
Here is my fourth year undergraduate work that focuses on using recurrent neural networks to handle memories in the form of histories in POMDPs to learn to generate macro actions for POMDP planning. The new architecture boosts the state-of-the-art MAGIC across all tested benchmarks.
This is my third year undergraduate work on designing efficient nerual network architecture for single image deraining. In particular, we recursively define transformer blocks to reuse weights to boost performance while keeping memory usages low. The performance surpassed many big models across different data sets at that time.
Other Experiences
Summer Research Scholarships
Online Attention in Social Media
Nov 2021 - Feb 2022
I worked with Prof. Lexing Xie on online attention markets and the effects of filter bubbles and echo chamber effects in social media. We investigated in different stochastic models to simulate the online interaction process and used the T-Recs simulator to perform experiments.
CSIRO Student Internships
Nov 2022 - Mar 2023
During the last year of my undergraduate, I worked under Dr Tirthankar Bandyopadhyay at CSIRO and built a simulation environment for robotic arms to interact with cluttered objects. Specifically, I used the Mujoco physics engine and OMPL library to achieve the goal.
More About Me
