Light-Field Controller for Robot Swarms
In this project, I designed a system for both charging and controlling a phototaxic (i.e. light-following) robot swarm using a light-field. The algorithm I designed can be run without direct controller-robot communication to cause the swarm to push boxes into a user-defined shaped. This project was supported by an NSERC USRA provided by the federal Canadian government. My paper was included in the conference proceedings for the International Conference on Robots and Automation (ICRA) 2019. This was my first scientific publication.
Open-Loop Collective Assembly Using a Light Field to Power and Control a Phototaxic Mini-Robot Swarm
Adam Bignell, Lily Lee, Richard Vaughan.
Proc. IEEE Int. Conf. on Robotics and Automation (ICRA'19), Montreal, Canada , May 2019
Comparison of ICTS and EPEA*
In my final project for an Intelligent Systems class, my group compared the relative efficacy of two advanced Multi-Agent Path Finding algorithms, the Increasing Cost Tree Search (ICTS), and Enhanced Partial Expansion A*. We tested our own implementations of these algorithms against MAPF instances of our own creation, and developed an automatic way to generate both open maps and mazes. Note that because we do not differentiate between a ‘wait’ action and a ‘move’ action, there are some solutions that appear superficially suboptimal (agents may walk back and forth while they wait for another agent to pass).
Detecting Conflict among Online Communities Using Relational Neural Nets
This was a final group project for a Machine Learning class I took in Fall 2018. In this project, we added a Relational Network to an existing neural architecture (Kumar et. al, 2018) for detecting conflict between Reddit communities. We presented our work at a class-wide poster session and won best project in the class as voted by attendees (which included various Machine Learning industry leaders from the Vancouver area). This project gave me hands-on experience training a novel neural network, applying it to real data, and extracting useful scientific results for presentation.
Image created by Nadia Ciobanu, one of my group members on the project.
Classifying Ironic Tweets with Sentiment Analysis and Machine Learning
This was a final group project for a Natural Language Processing class I took in Fall 2018. Here, we classified tweets as either ironic or sincere based on a suite of features that we developed by hand. Some of these included contrast between the tweet text sentiment and emoji sentiment (“I just love being late 😤“), presence of highly ironic phrases "(“oh how I”), and total emotional tension of the text (many happy words alongside sad words). While the project did not require a paper, and required that the code remain private, we did produce a poster of our results.
The emotional valence of various popular emojis. Each emoji was assigned a Positive, Negative, and Neutral score, each between 0 and 1 such that the sum of all three scores sum to 1.
Dynamic Worksite Relocation for Foraging Robot Swarms
Done as a term project for a graduate level robotics class in Spring 2018. This class gave me hands-on experience with robot simulators, and helped me to think in terms of swarm dynamics and emergent behaviour. I designed an extension of the relocating bucket brigade algorithm for robot swarms (see Lein, 2009). Under my algorithm, the robots count how many resources are at a given location, and relocate their worksites towards locations with higher resources. Unfortunately, my results did not significantly improve the previous algorithm: this was likely because of the high degree of robot-to-robot interference that occurs once the robots have relocated most of the resources in the direction of the sink.
Note that the video above is without my extension, and was an exercise on if I could reimplement the bare bucket-brigade algorithm successfully. The video does however show the algorithm running to completion. Highly recommend 2x playback speed.