My Work

I am an Early Research Career Fellow at Technology, CSIRO, Australia, working in the Human-Robot Interaction team. My research focuses on deep reinforcement learning, reward shaping, and robotic manipulation, with an emphasis on helping learning agents acquire robust behavior from limited interaction.

During my Ph.D., I worked on adaptive potential functions for reinforcement learning, exploring how task guidance and value estimation can help agents learn more efficiently. My current work studies how robotic imitation learning can generalize better under distribution shifts. In particular, I investigate how policies can reduce causal confusion in observations by focusing on task-relevant information and avoiding spurious correlations in demonstrations. This aims to make learned robot policies more reliable when the environment changes, such as under different backgrounds, distractors, object appearances, or task conditions.