Technological advancements have led to a proliferation of robots using machine learning systems to assist humans in a wide range of tasks. However, we are still far from accurate, reliable, and resource-efficient operations of these systems. Despite the strengths of convolutional neural networks (CNNs) for object recognition, these discriminative techniques have several shortcomings that leave them vulnerable to exploitation from adversaries. In addition, the computational cost incurred to train these discriminative models can be quite significance. Discriminative-generative approaches offers a promising avenue for robust perception and action. Such methods combine inference by deep learning with sampling and probabilistic inference models to achieve robust and adaptive understanding. The focus is now on implementing a computationally efficient generative inference stage that can achieve real-time results in an energy efficient manner. In this talk, I will present our work on Generative Robust Inference and Perception (GRIP), a discriminative-generative approach for pose estimation that offers high accuracy especially in unstructured and adversarial environments. I will then describe how we have designed an all-hardware implementation of this algorithm to obtain real-time performance with high energy-efficiency.
Department of Computer Science