Josselin Somerville Roberts's repositories
BERT-Multitask-learning
Multitask-learning of a BERT backbone. Allows to easily train a BERT model with state-of-the-art method such as PCGrad, Gradient Vaccine, PALs, Scheduling, Class imbalance handling and many optimizations
ReachBot-Predictor
Repository of the paper: A Skeleton-based Approach For Rock Crack Detection Towards A Climbing Robot Application
UltimateTicTacToe-RL
Ultimate Tic-Tac-Toe game and agents. The following agents are implemented: random, Monte-Carlo, Min-Max, DQN
hackathon-spot
Spot My good boy is a personal assistant that you can interact with and that can help you find object you lost in the room
SpeechBubbleSubtitles
The project aims to generate speech bubble automatically on a video only using the video and a subtitles file.
Autonomous-Parking-Lot-Sweeper
This project aims to optimize the path of an autonomous vehicle that sweeps parking lot. The aim is to maximize the covered area while minimizing the time to clean it.
BlackJackCountingEnv
Reinforcement learning agent that learns the basic strategy (most optimal choice in every situation) and then how to adapt the betting and playing strategy based on the count.
Echo-s-Adventures
An interactive world coded from scratch in C++ using OpenGL.
H-index-predictor
Using a co-citation network and abstracts to predict the H-index of a researcher.
JossCustomCommands
Custom commands for Ubuntu used for my work. This includes commands to start and stop with a single command some AWS EC2 instances
JosselinSomervilleRoberts.github.io
Personal Website
JossPythonToolbox
Python library to provide basic useful functions. This includes debugging printing functions, logging and AWS management.
Procedural-Dialog-Generation
Procedural dialog generation for NPC in games (with Ubisoft)
ProjectOcean
Fast wave generation in OpenGL with various rendering effects: BRDF, reflection, foam generation, buoyancy approximation, ...
robotsCGShop2021
Optimization challenge. The goal consisted of determining paths for robots moving on the same map to go from a starting to ending position without obstructing other robots
Single-life-legged-robot-learning
CS224R Final Project: Deep Reinforcement Learning
SSR-Renderer
Screen Space Reflection Renderer coded from scratch in C++
Starcraft-Bot
Using Behaviour Trees to create an AI for Starcraft with BWAPI.
arxivscraper
A python module to scrape arxiv.org for a date range and category
garage-improved
A toolkit for reproducible reinforcement learning research.
Magic-Arena
Multiplayer Arena game coded from scratch in C++
minnow
CS 144 networking lab
TabularRL-Experiments
Experiments on RL exploration on tabular environments to generalize concepts in hte Madrona engine
VLM-image2structure-eval
Playground for image2structure scenarios to evalute VLM on.