[month] [year]

Gulshan Kumar – Dual Degree CSE

Gulshan Kumar received his MS Dual Degree in Computer Science and Engineering (CSE). His research work was supervised by Prof. Madhava Krishna. Here’s a summary of  his research work on Goal-Conditioned Exploration for Object Goal Navigation:

Autonomous navigation is a critical task for building intelligent embodied AI agents. For a successful visual navigation algorithm, building spatial representations for efficient exploration and exploiting structural priors is necessary irrespective of the navigation task. Imagine you are in a house which you have never seen before and you are given the task of ”Finding a sink” in that house. While there can be multiple directions to move, most of us would choose the path that would take us to the Dining Room, where the sink is highly likely to occur in the Kitchen. This is because as a human we use strong structural priors to navigate in an unseen environment. This work incorporates these semantic priors on the relative arrangement of objects in a scene as part of our scene understanding module for solving the navigation task effectively. In this thesis, we address the highly challenging problem of object goal navigation in which the agent is tasked to reach any instance of the specified goal category within a defined number of time-steps. The agent, in an unseen environment, has to perceive its surroundings to identify and navigate towards potential regions where the specified goal category can occur. Rather than developing goal driven exploration policies, we aim to adapt the existing exploration policies that maximize scene coverage to be goal-conditioned. Thus, we propose a standalone scene understanding module to identify potential regions where the goal occurs. We also propose Goal-Conditioned Exploration (GCExp), an algorithm that entails the integration of our novel scene understanding module with any existing exploration policy. We test our solution in photo-realistic simulation environments using state of-the-art exploration policy, Active Neural Slam and show improved performance over the same on every evaluation metric. Additionally, we show a marginal improvement on the newly proposed evaluation metric, ”Average Categorywise Success”, to address the shortcomings of the existing metrics.