Visual simultaneous localization and mapping (SLAM) is an emerging technology that enables low-power devices with a single camera to perform robotic navigation. Most visual SLAM algorithms are tuned for images produced through the image sensor processing (ISP) pipeline optimized for highly aesthetic photography. We investigate the feasibility of varying sensor quantization on RAW images directly from the sensor to save energy for visual SLAM. An 88% energy savings has been achieved by decreasing quantization bit level to five bits. We also introduce a gradient-based quantization scheme that increases energy savings. This work opens a new direction in energy-efficient image sensing for SLAM.
Heat poses a major health risk that particularly affects cyber-physical infrastructures in cities. Heat-sensing maps can be used to reduce overexposure risks to humans. In this research project, we analyze how a robot equipped with a thermal camera can efficiently acquire a map of the surrounding area, localize its position within the map, and autonomously navigate this map to perform heat-sensing measurements. We evaluate how different SLAM algorithms perform using thermal imagery and determine optimal navigation and sensing policies to ensure these robots can efficiently scan and detect heat hazards and changing environmental conditions.