Papers
8
Total Citations
71
H-Index
5
About
Ashley Hill’s research sits at the intersection of robotics, reinforcement learning, and state representation learning, with a focus on enabling real-world, off-road mobile robots to navigate safely and efficiently. Her most influential work, “Decoupling feature extraction from policy learning” (24 citations), challenges the end-to-end learning paradigm by demonstrating how state representation learning can dramatically improve sample efficiency when controlling real robots from vision. This foundational insight is complemented by her co-creation of the S-RL Toolbox (20 citations), a widely used open-source benchmark that provides environments, datasets, and evaluation metrics for state representation learning—a resource that has become a standard for researchers in the field. Hill has also pioneered novel approaches for online control gain tuning, using neural networks and the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to adapt robot behavior in real time under poor grip conditions. Her work on velocity fluctuation and path tracking in off-road settings (8 and 6 citations) directly addresses the practical challenge of balancing speed with safety. Through her development of gradient-based feature importance methods for neural networks, Hill has contributed interpretability tools that help roboticists understand and trust their learned controllers. Her research is characterized by a rare combination of theoretical depth and hands-on robotic implementation, making her a leading voice in bringing reinforcement learning from simulation to rugged, real-world terrains.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8