Fast, lightweight autonomy through an unknown cluttered environment: Distribution statement: A — Approved for public release; distribution unlimited

The mission of clearing buildings remains among the most dangerous operations performed by military teams and first responders. Personnel searching for hidden combatants or trapped victims must move quickly through hazardous and unfamiliar spaces, facing life-or-death situations around each corner. A smart robotic scout could help perform rapid reconnaissance while search teams safely monitor progress from the outside, but walls, doors, and cluttered spaces make it difficult for operators to pilot the vehicles remotely. Towards this vision, the Defense Advanced Research Projects Agency (DARPA) Fast Lightweight Autonomy (FLA) program aims to create a new class of navigation, perception, planning, and control algorithms which enables autonomous, high-speed flight through unknown, cluttered environments. Integrating this new technology with small quad-rotor unmanned aerial vehicles (UAVs), the program will demonstrate aggressive, agile flight where the vehicle will autonomously sense and maneuver through unknown indoor/outdoor environments without external communications or GPS at speeds up to 20 meters/second. The Draper/MIT team is working to create advanced navigation, perception, planning, and control algorithms running on-board a custom, tightly integrated embedded computing and sensing avionics platform to accomplish the FLA program goals. Our algorithms must be exceedingly efficient, effective, and low-latency to enable this real-time agile perception and planning capability through cluttered, unknown environments. To support this capability we rely extensively on high quality IMU and monocular camera data to enable accurate GPS-denied navigation and perception of the immediate environment for path planning and collision avoidance maneuvering. The software developed under this program is planned for eventual release as open source by DARPA, to further advance the state-of-the-art in autonomy. This technology has applicability to not only UAVs, but to any autonomous robotic platform exploring unknown complex environments where moving quickly yet avoiding hazardous obstacles is critical. Additionally, using indoor capable (i.e. small) quad-rotor UAVs as the development platform guarantees that the resulting avionics system will be low Size, Weight, and Power (SWaP), and therefore ideal for a variety of autonomous robotic (air, land, sea, and space) applications. This paper describes the high-level system architecture and algorithmic approaches being utilized by the Draper/MIT team, as well as reports on recent field testing results.