Bryan Lim

I am a PhD student at the Adaptive and Intelligent Robotics Lab, part of the Department of Computing at Imperial College London, where I work on robotics and machine learning. My PhD advisor is Dr. Antoine Cully.

Previously, I completed my MEng year at MIT, where I worked with Prof. Alberto Rodriguez and Prof. Sangbae Kim. I have an MEng in Mechanical Engineering from Imperial College London.

 /   /   /   /  Google Scholar

Research

I work at intersection of machine learning and robotics. My research aims to make AI systems more reliable and useful by making them more robust, creative, general-purpose and have the ability to continuously improve. I believe open-ended learning systems which can continously learn and generate intersting things can be a solution to this. To enable such systems, I study Quality-Diversity algorithms which encourages novelty and diversity as a key component in the process. More specifically, I am interested in the synergy between deep models (i.e. generative/foundation models) and Quality-Diversity search which can lead to self-improvement.

News

• [5/2024] In-context QD: our work on many-shot in-context learning with high-quality and diverse (Quality-Diverse) examples was accepted at ALIFE 2024!
• [12/2023] Our work on accounting for policy reproducibility when evaluating RL algorithms was accepted at AAAI!
• [6/2023] Check out our work on using RL for behavioural diversity of a quadruped, including dynamic and agile three-legged locomotion behaviour. Video here! Interestingly similar athletic intelligence and adaptability to that of animals!
• [4/2023] GDAQD accepted at ALIFE 2023!
• [3/2023] Two papers and a poster accepted at GECCO 2023. Going to be a fun summer week in Lisbon!
• [1/2023] Our work studying QD-based neuroevolution and unsupervised deep RL methods was accepted at ICLR 2023! Grateful to be able to work with such a great team of collaborators on this one!
• [1/2023] One paper accepted at ICRA 2023. See you in London!
• [12/2022] Will be presenting our work on scaling model-based QD with gradient updates at the NeurIPS Deep Reinforcement Learning Workshop!
• [7/2022] We officially released QDax, an open-source library in JAX for Quality-Diversity and Population-based algorithms. Check it out!
• [3/2022] I was awarded the Imperial College Robotics Forum - Amazon PhD Prize for Outstanding Achievement in Robotics!
• [3/2022] Will be presenting 2 papers at the Agent Learning in Open-Endedness Workshop at ICLR 2022. We will present QDax in a spotlight talk!
• [3/2022] 1 paper and 1 poster accepted to GECCO 2022! Excited to be heading back to Boston!
• [1/2022] DA-QD accepted at ICRA 2022! My first paper since strating the PhD!

Conference Papers

Large Language Models as In-context AI Generators for Quality-Diversity

Enhancing MAP-Elites with Multiple Parallel Evolution Strategies

Beyond Expected Return: Accounting for Policy Reproducibility when Evaluating Reinforcement Learning Algorithms

Quality-Diversity Optimization on a Physical Robot through Dynamics-Aware and Reset-free Learning

Don't Bet on Luck Alone: Enhancing Behavioral Reproducibility of Quality-Diversity Solutions in Uncertain Domains

Understanding the Synergies between Quality-Diversity and Deep Reinforcement Learning

Efficient Exploration using Model-Based Quality-Diversity with Gradients

Neuroevolution is a Competitive Alternative to Reinforcement Learning for Skill Discovery

Efficient Learning of Locomotion Skills through the Discovery of Diverse Environmental Trajectory Generator Priors

Accelerated Quality-Diversity through Massive Parallelism

Learning to Walk Autonomously via Reset-Free Quality-Diversity

Dynamics-Aware Quality-Diversity for Efficient Learning of Skill Repertoires

Tactile object pose estimation from the first touch with geometric contact rendering

Robust Autonomous Navigation of a Small-Scale Quadruped Robot in Real-World Environments

Vision aided dynamic exploration of unstructured terrain with a small-scale quadruped robot

Journal Papers

QDax: A Library for Quality-Diversity and Population-based Algorithms with Hardware Acceleration

Online Damage Recovery for Physical Robots with Hierarchical Quality-Diversity

Workshop Papers

Benchmarking Quality-Diversity Algorithms on Neuroevolution for Reinforcement Learning

Teaching

• C70050 Introduction to Machine Learning, TA
• C70028 Reinforcement Learning, TA
• C70067 Robot Learning and Control, TA

Design and source code from Jon Barron's website and Leonid Keselman's Jekyll fork