Natural Language Processing for Robotic Navigation

Natural Language Processing for Robotic Navigation Through Unkown Environments

Presenter: Olivia Piazza

Presenter Status: Graduate student

Department: Engineering

Screenshot URL: https://drive.google.com/uc?id=1l0RFQdLbqablVxXOPdMfagN1bGh-ebRq

Abstract:
In this project, we intend to present a navigation system for goal-directed exploration in unfamiliar environments on a physical robot using quasi-ambiguous verbal commands. We introduce a novel construct for human-machine interfacing (HMI), allowing the human operator to provide directions to the robot in a similar way a human might provide direction to another human to reach a specific location. Our model will treat the instruction-environment relation as a trajectory optimization problem, where the agent needs to learn the best path within the environment to navigate to the goal, through language grounding within the visual environment and instruction interpretation using pragmatic speech frameworks in a reinforced learning context. Once the agent is trained, timed results will be compared to the time required to navigate to the goal through random exploration and reinforced learning, demonstrating the advantage of our novel HMI over previous methods.