Jost Tobias Springenberg
University of Freiburg, Google (United States), Google (United Kingdom), Google DeepMind (United Kingdom)
Papers
26
Total Citations
1,216
H-Index
14
About
Jost Tobias Springenberg is a prominent researcher at the intersection of deep reinforcement learning, robot control, and autonomous skill acquisition. His work has made substantial contributions to some of the most challenging problems in modern AI, particularly enabling robots and agents to learn complex behaviors with minimal human supervision. Springenberg's most-cited contributions center on transfer learning and navigation, with his work on successor features for robot navigation (283 citations) demonstrating how agents can generalize across similar environments without explicit mapping or localization. His research on skill embedding spaces (190 citations) and the Scheduled Auxiliary Control framework, SAC-X (155 citations), tackled the notoriously difficult sparse reward problem, showing that agents could learn rich, reusable behaviors essentially from scratch. Early in his career, he also advanced multimodal deep learning for RGB-D object recognition (133 citations), underscoring his breadth across perception and control. More recently, Springenberg contributed to the landmark "Generalist Agent" (Gato) project, exploring a single unified policy capable of multi-task, multi-embodiment behavior. His work on offline reinforcement learning and robustness to model misspecification further reflects a consistent drive toward making reinforcement learning practical and reliable in real-world robotics settings.
Research Focus
Key Achievements
Top Papers
- 1
- 2Learning an Embedding Space for Transferable Robot Skills190 citations · 2018
- 3Learning by Playing - Solving Sparse Reward Tasks from Scratch155 citations · 2018
- 4Multimodal deep learning for robust RGB-D object recognition133 citations · 2015
- 5A Generalist Agent66 citations · 2022
- 6
- 7
- 8
- 9
- 10