Daan Geijs

Papers

Total Citations

H-Index

About

Daan Geijs is a researcher in reinforcement learning, with a particular focus on representation learning and robotics. His most notable contribution is the development of reward-shaped priors for learning low-dimensional state representations, a method that significantly reduces the data and computational burden required for training policies directly from high-dimensional sensory inputs. This work, published in 2021 and garnering 4 citations, addresses a critical bottleneck in end-to-end reinforcement learning: the need to process and store vast amounts of observation data. By shaping the reward signal to guide the learning of compact, task-relevant state representations, Geijs’s approach enables more efficient policy learning without extensive feature engineering. His research sits at the intersection of reinforcement learning, robotics, and representation learning, aiming to make autonomous systems more sample-efficient and scalable. While early in his career, Geijs’s work on reward-shaped priors offers a promising pathway toward practical, data-efficient learning in complex robotic tasks.