Edge computing near the sensor is preferred over the cloud due to privacy or latency concerns for a wide range of applications including robotics/drones, self-driving cars, smart Internet of Things, and portable/wearable electronics.  However, at the sensor there are often stringent constraints on energy consumption and cost in addition to throughput and accuracy requirements. In this talk, we will describe how joint algorithm and hardware design can be used to reduce energy consumption while delivering real-time and robust performance for applications including deep learning, computer vision, autonomous navigation and video/image processing.  We will show how energy-efficient techniques that exploit correlation and sparsity to reduce compute, data movement and storage costs can be applied to various AI tasks including object detection, image classification, depth estimation, super-resolution, localization and mapping.