Happy to finally announce the first full release of QDax! We’re super excited about the use of massive parallelization and hardware acceleration for Quality-Diversity algorithms that we decided to develop a library for this!
Bryan Lim
74 posts
Bryan Lim
@bryanlimwt
Building agents and robots. Prev @autodesk ai lab, PhD @imperialcollege, MS @imperialcollege @MIT. AI/ML/Robotics.
Joined February 2021
- Interested in how to train diverse population of agents more efficiently? Check our our new paper being presented at the @NeurIPSConf DeepRL workshop today! Paper: arxiv.org/abs/2211.12610 This is work done with amazing collaborators @MFlageat (Project Co-lead) and @acully.
- Very happy to share that I passed my PhD viva yesterday! Extremely grateful to @CULLYAntoine who has been a great advisor and mentor throughout this journey. Thank you for giving me this opportunity of a lifetime!
- Excited for the @aloeworkshop at @iclr_conf tomorrow! We will be presenting our work on accelerating Quality-Diversity (QD) algorithms (Spotlight Talk at 4.15 pm GMT+1) which makes QD algorithms more accessible and which we hope will help in scaling open-ended learning.
GIF - Dynamics-Aware Quality-Diversity (DA-QD) We explore combining QD algorithms for skill-discovery with learnt dynamics models to get a ~20x increase in sample-efficiency! Excited to share my first paper since starting my PhD w/ @CULLYAntoine Paper: arxiv.org/abs/2109.08522
GIF - Excited to share our new paper, Learning to Walk Autonomously via Reset-Free Quality-Diversity (RF-QD), an effort towards more autonomous robot learning! RF-QD enables the discovery of diverse skill repertoires without the need for resets when deployed in cluttered environments.
00:00 - Our new work aims to increase the accessibility of QD algorithms. Through massive parallelism, large and diverse behavioral repertoires can be obtained in minutes instead of hours/days with negligible effect on the final performance.Excited to share “Accelerated Quality-Diversity for Robotics through Massive Parallelism”. We introduce QDax which makes QD 100 times faster by using massive batch sizes on GPU/TPU. With @bryanlimwt @allardmaxime079 and @GrillottiLuca arXiv: arxiv.org/abs/2202.01258 (more below)
- Happy to share that Dynamics-Aware Quality-Diversity (DA-QD) has been accepted at #ICRA2022! Work done with @GrillottiLuca @CULLYAntoineDynamics-Aware Quality-Diversity (DA-QD) We explore combining QD algorithms for skill-discovery with learnt dynamics models to get a ~20x increase in sample-efficiency! Excited to share my first paper since starting my PhD w/ @CULLYAntoine Paper: arxiv.org/abs/2109.08522GIF
- Excited to be presenting this year’s edition of Evolutionary Reinforcement Learning (EvoRL) @GeccoConf 🦎🦾 with @MFlageat @CULLYAntoine, which will be a tutorial instead of a workshop! 🧵
- 💡⚙️ Ideas and inventions are rarely generated and created in isolation 🏝️ We explore using few/many-shot prompting of language models with quality-diverse examples provided from QD algorithms for solution generation for QD problems! Work done with @MFlageat @CULLYAntoine
- Ever wondered how far we can push the dynamic capabilities of robots? Here’s a snippet of some ongoing stuff @NathanAmar10 and I are working on!Excited to share a project I’ve been working on with @bryanlimwt and @CULLYAntoine. We push the boundaries of what Reinforcement Learning can do for legged robots by achieving dynamic three legged locomotion 🤖🐕. More coming soon! 🚀 Extended Video: youtube.com/watch?v=xb2gOo…
00:00 - Had a great time at @ALifeConf this week presenting In-context Quality-Diversity! Feel free to reach out if any of this is interesting #ALIFE2024
- Replying to @bryanlimwtBack home in Malaysia so instead of post-viva drinks, celebrated with some Nasi Kandar (curry rice)!
- I’m at #ICRA2023 this week to present our work on learning trajectory generator priors for locomotion! Happy to meet up for a chat or come say hi at our poster session on Thursday (3pm)!
