Papers
9
Total Citations
51
H-Index
4
About
Sarah Bechtle is a robotics and machine learning researcher whose work sits at the intersection of model-based reinforcement learning, inverse reinforcement learning, and robot cognition. Her research addresses some of the field's most pressing challenges: enabling robots to learn efficiently from limited data, infer reward functions from demonstrations, and develop rich internal representations of their bodies and environments. Among her most influential contributions is "Curious iLQR" (2019, 16 citations), which pioneered the integration of Bayesian curiosity-driven exploration into model-based reinforcement learning, offering a principled approach to uncertainty resolution during robot control. Her work on model-based inverse reinforcement learning from visual demonstrations (2020, 8 citations) advanced the scalability of reward learning to real-world manipulation tasks. Bechtle has also made notable strides in developmental robotics, exploring how humanoid robots can acquire a sense of agency and object permanence — concepts rooted in developmental psychology and neuroscience. More recently, her research has expanded into large-scale learning, demonstrating that offline actor-critic methods can follow transformer-era scaling laws (2024), and that iterative reinforcement learning on real robots can master physically diverse manipulation tasks. Her body of work reflects a sustained commitment to bridging cognitive science and practical robot learning.
Research Focus
Key Achievements
Top Papers
- 1Curious iLQR: Resolving Uncertainty in Model-based RL16 citations · 2019
- 2On the sense of agency and of object permanence in robots11 citations · 2016
- 3Model-Based Inverse Reinforcement Learning from Visual Demonstrations8 citations · 2020
- 4
- 5
- 6
- 7
- 8
- 9Offline Actor-Critic Reinforcement Learning Scales to Large Models2 citations · 2024