PhD Proposal: Parsimonious AI: Software-Hardware Synergy For Robot Autonomy
Although autonomous agents are inherently built for a myriad of applications, their perceptual systems are designed with a single line-of-thought - perception delivers a 3D representation of the scene. The utilization of these traditional methods on most autonomous agents (like drones) is highly inefficient as these algorithms are generic and not parsimonious. In stark contrast, the perceptual systems in biological beings have evolved to be highly efficient based on their natural habitat as well as their day-to-day tasks.We draw inspiration from nature to build a minimalist cognitive framework for robots at scales that were never thought possible before. We propose a novel Parsimonious AI framework for mobile robots to solve a class of perception and navigational tasks like traversing through dynamic unstructured environments and segmenting never-seen objects. We utilize the fact that computing is only a small aspect of a robot. We re-imagine the robot from the ground-up based on a class of tasks to be accomplished. This leads to a set of tight constraints that aid in efficiently solving the problem of autonomy.
Dr. Yiannis Aloimonos Dr. Christopher Metzler Dr. Guido de Croon (TU Delft)Dr. Cornelia FermüllerDr. Nitin J. Sanket