Project

Controlling Autonomous Agents in Dynamic Environments

PI(s): Nakul Gopalan
UNIVERSITY: Arizona State University

CODE: 120.ASU.GOP.26.Y3
STATUS: Continuing

University Logo

DESCRIPTION:

A significant challenge in problem domains like agriculture and construction is that the environment is unstructured and presents significant variety of hazards on site. Such lack of structure and hazard has prevented the progress of automation in these fields. In this work we attempt to learn safe policies for large vehicles to navigate and operate in unstructured environments with risks such as human movements and presence of hazards such as ditches and piles of dirt.

Our objectives are to learn navigation and manipulation policies within a simulator for a large vehicle in unstructured environments. We propose to learn these policies on a skid-steer, which is a proxy for large vehicles such as bulldozers, and tractors. These policies would be learned with visual inputs within a simulator. Given we do not have a simulator for such environments and a skid-steer we will create a novel simulation environment and add dynamic human elements along with hazards to replicate construction and hazard sites. A significant challenge here is also that humans might move differently in unstructured and hazardous environments.