Papers
27
Total Citations
2,579
H-Index
16
About
Alexander Herzog is a leading roboticist whose work spans the critical intersection of manipulation, locomotion, and learning. His research has fundamentally advanced how robots interact with the physical world, from dexterous grasping to dynamic whole-body control. Herzog’s most impactful contributions lie in scaling deep reinforcement learning for real-world robotic tasks. His seminal work on QT-Opt (575 citations) pioneered a scalable framework for learning vision-based manipulation, while the RT-1 (512 citations) and RT-2 (267 citations) models revolutionized the field by integrating large-scale web knowledge into robotic control, enabling unprecedented generalization and semantic reasoning. Earlier in his career, Herzog made foundational contributions to humanoid robotics, developing hierarchical inverse dynamics controllers for torque-controlled robots—demonstrated through balancing experiments (143 citations) and momentum control (254 citations). He also advanced grasp planning through shape-template learning (113 citations) and multi-contact motion generation (81 citations). Herzog’s work has been instrumental in bridging model-based control with data-driven learning, establishing him as a key figure in the modern robotics landscape.
Research Focus
Key Achievements
Top Papers
- 1
- 2RT-1: Robotics Transformer for Real-World Control at Scale512 citations · 2023
- 3RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control267 citations · 2023
- 4
- 5
- 6Learning of grasp selection based on shape-templates113 citations · 2013
- 7Template-based learning of grasp selection100 citations · 2012
- 8Trajectory generation for multi-contact momentum control88 citations · 2015
- 9Structured contact force optimization for kino-dynamic motion generation81 citations · 2016
- 10