Adrien Ecoffet

Papers

Total Citations

490

H-Index

About

Adrien Ecoffet is a leading researcher in artificial intelligence, best known for his groundbreaking work on exploration in reinforcement learning (RL). He is the principal architect of the **Go-Explore** family of algorithms, which directly tackle the fundamental challenge of learning in environments with sparse or deceptive rewards. His seminal paper, *"Go-Explore: a New Approach for Hard-Exploration Problems"* (2019, 228 citations), introduced a paradigm shift by first returning to previously visited states before exploring, a simple yet powerful idea that achieved unprecedented scores on notoriously difficult Atari games like Montezuma's Revenge and Pitfall. This work was further refined in *"First return, then explore"* (212 citations), solidifying the approach as a cornerstone of modern exploration research. More recently, Ecoffet contributed to **Video PreTraining (VPT)** (2022, 50 citations), a novel method for learning to act by watching unlabeled online videos, demonstrating how internet-scale data can bootstrap generalist agents for sequential decision-making. His contributions have fundamentally reshaped how the RL community approaches hard-exploration problems, bridging the gap between toy environments and real-world complexity.