In robotics, motion planning refers to the problem of navigating from start to goal positions while avoiding collisions with the environment. Model-based techniques use knowledge of obstacle properties and robot dynamics to construct paths, while neural motion planners learn from a dataset and produce new solutions through generative modeling. These methods balance tradeoffs between optimality and computation time, with an ultimate goal of generalization to unseen environments. We explore different classes of generative models to leverage the advantages of distinct motion planning techniques. Specifically, we seek to improve the success of neural motion planners by incorporating aspects of model-based methods.

I am excited to refine my ability to plan long-term experiments, communicate to a technical audience, and contribute to the wider field of robotics.