Navion: A 2mW Fully Integrated Real-Time Visual-Inertial Odometry\n Accelerator for Autonomous Navigation of Nano Drones

Abstract

This paper presents Navion, an energy-efficient accelerator for\nvisual-inertial odometry (VIO) that enables autonomous navigation of\nminiaturized robots (e.g., nano drones), and virtual/augmented reality on\nportable devices. The chip uses inertial measurements and mono/stereo images to\nestimate the drone's trajectory and a 3D map of the environment. This estimate\nis obtained by running a state-of-the-art VIO algorithm based on non-linear\nfactor graph optimization, which requires large irregularly structured memories\nand heterogeneous computation flow. To reduce the energy consumption and\nfootprint, the entire VIO system is fully integrated on chip to eliminate\ncostly off-chip processing and storage. This work uses compression and exploits\nboth structured and unstructured sparsity to reduce on-chip memory size by\n4.1$\\times$. Parallelism is used under tight area constraints to increase\nthroughput by 43%. The chip is fabricated in 65nm CMOS, and can process\n752$\\times$480 stereo images from EuRoC dataset in real-time at 20 frames per\nsecond (fps) consuming only an average power of 2mW. At its peak performance,\nNavion can process stereo images at up to 171 fps and inertial measurements at\nup to 52 kHz, while consuming an average of 24mW. The chip is configurable to\nmaximize accuracy, throughput and energy-efficiency trade-offs and to adapt to\ndifferent environments. To the best of our knowledge, this is the first fully\nintegrated VIO system in an ASIC.\n