Student: Anirudhudu Duddilla
Committee: Dr. Zhe Xu
Abstract:
Existing Object Detection methods in autonomous driving are trained on the data sets that are biased towards good weather conditions which often fail in rare weather conditions. Specifically, we modeled the agent’s behavior as an uncertain Partially Observable Markov Decision Process (POMDP), a widely used model for decision-making under uncertainty and imperfect information by expressing the task requirements as linear temporal logic (LTL) specifications, which has been widely used for specifying agents’ behavior. Computed a policy in the underlying (uncertain) POMDP that visits an ordered set of N locations such that the probability of obtaining a correct label for each object is greater than threshold \gamma(ensures agent selects the objects with highest probability), while ensuring safety and minimizing time spent. Case study has been done to demonstrate the proposed approach.
Zoom Room: https://asu.zoom.us/j/7176999347
Presentation Time: 12:00-1:00 PM (Arizona Time)
