Efficient Visual Avoidance and Control for UAVs with Embedded GPUs

Explore a comprehensive conference talk on efficient visual avoidance and control for UAVs using embedded GPUs. Delve into the challenges of indoor and outdoor navigation for small UAVs, particularly quadrotors, and discover how Nvidia's Jetson TX1 embedded GPU compute module and high-speed vision systems can revolutionize onboard navigation. Learn about a compact quadrotor capable of tight maneuvers, safe GPS-denied indoor hover, onboard mapping, sense-and-avoid capabilities, precision landing, and HD streaming. Gain insights into the Artemis micro-aerial-vehicle research project, which aims to make autonomous obstacle avoidance and GPS-denied navigation possible for small multi-rotor craft. Examine the complete navigation stack, including depth extraction, odometry, mapping, planning, and control, all running onboard at a fraction of the weight and power consumption of traditional CPU architectures. Understand the vehicle platform, sensing suite, visual-inertial odometry, state estimation, global planning, nonlinear optimal control, operator interfaces, and multi-sensor fusion techniques employed in this cutting-edge UAV system.

Intro
STATE OF THE INDUSTRY
PROJECT ARTEMIS - A HISTORY
DESIGN GOALS
VEHICLE PLATFORM - Artemis MAVS
VEHICLE SPECS
TEGRA X1 COMPUTE PLATFORM
THE NAVIGATION PROBLEM
SENSING SUITE Framerate
VISUAL-INERTIAL ODOMETRY
VISUAL-INERTIAL STATE ESTIMATION
GLOBAL PLANNING
NONLINEAR OPTIMAL CONTROL
OPERATOR INTERFACE
SOFTWARE FRAMEWORK
COMPLETE NAVIGATION PIPELINE
MULTI-SENSOR FUSION