John B. Lanier

Papers

Total Citations

H-Index

About

John B. Lanier is a researcher advancing reinforcement learning (RL) in complex, sparse-reward environments. His work centers on overcoming the fundamental challenge of learning from minimal feedback, particularly in robotic manipulation and multi-step tasks. Lanier’s major contribution is the development of "Curiosity-Driven Multi-Criteria Hindsight Experience Replay" (2019), a method that integrates intrinsic curiosity with hindsight learning to enable agents to tackle problems where standard hindsight methods fail—such as stacking multiple blocks with a simulated robot arm. By combining multiple reward criteria with exploration bonuses, his approach significantly improves sample efficiency and task success in high-dimensional, long-horizon settings. Though his most-cited paper has garnered 11 citations, its impact is growing as the field increasingly confronts the limitations of existing RL techniques. Lanier’s work is notable for bridging theoretical curiosity-driven exploration with practical, multi-objective learning, offering a pathway toward more autonomous and capable agents. For students and researchers, his research highlights a critical frontier: making RL robust enough for real-world tasks where rewards are sparse and goals are complex.