Embedded Files
Olin College IARC at Georgia Tech, 2019
IARC is the longest running collegiate aerial robotics challenge in the world.
It is intended to move the state-of-the-art in aerial robotics forward through the creation of significant and useful mission challenges that are considered 'impossible' at the time that they are proposed.
In 2018-2019, Eric Miller, Jaimie Cho, Max Schommer, and I founded Olin's IARC project team and lead a group of underclassman to develop a robotic drone swarm that could challenge Mission 8. [Fig. 3]
We successfully competed with our swarm, and won Best Presentation at Georgia Tech in 2019.
Mission 8
Mission 8
The goal of Mission 8 was for a single human was to coordinate with a team of up to four fully autonomous aerial robots to reassemble fragments of a QR code, while under fire from hostile aerial robots.
Navigation commands to your team of robots must be communicated through either gesture or vocal commands.
Platform
Platform
Our swarm was composed of four modified Parrot Bebop 2 drones. Bebops were chosen for their relative durability, accurate indoor height sensing, and stabilized camera image.
In addition to the included camera, downwards facing ultrasonic, 3-axis, gyroscope, magnetometer, accelerometer, and GPS, we attached...
three additional angled ultrasonic sensors to improve pitch angle sensing
a remote wifi-powered RC receiver
a remotely operable physical "hard" emergency stop (required)
four mesh propeller guards (required)
Software Design
Software Design
Layered Autonomy Stacks
Layered Autonomy Stacks
Our team split into four task-driven subteams for development.
Perception (QR detection and stitching, scene understanding, enemy vehicle detector)
Planning and Controls (inflight behavior arbiter, state estimation, inflight obstacle avoidance, flight simulators)
Infrastructure (Wifi interface, drivers, hardware)
Testing (Integration testing, flight characterization)
I led the Planning and Controls subteam.
Behavior Arbiter
Behavior Arbiter
The behavior arbiter takes into account commands received from the speech processing module, as well as sensor input, when planning the next autonomous movement.
It uses the artificial potential field algorithm to bias away from obstacles, and calculates the covariance of its current position to determine relative error bounds for acceptable flight paths. We used ROS to communicate between nodes in the software stack.
We tested the behavior arbiter using two custom built simulators...
Fast, 2D simulator for general behavior validation
High fidelity physics based flight simulation in Gazebo
Sound Detection
Sound Detection
The human participant wore a headset with a microphone. The microphone transmitted directly back to the central command computer via wifi, where a speech processing module interpreted words and mapped them to a dictionary of known commands.
Testing, or Lessons Learned
Testing, or Lessons Learned
Propeller guards cut thrust more drastically than anticipated- at 100% thrust, the Bebops could only fly about 1m off the ground.
Wifi signals at competition were unreliable due to interference from other competitors' signals. Amplifiers or a fallback means of communication is critical.
Our speech detection system needed to be more forgiving of "misheard" commands in a noisy environment.
Accurate visual odometry was critical to remaining within bounds and fulfilling competition objectives.
Project Artifacts
Project Artifacts
Figure 1: Olin IARC team poster session
IARC_Poster.pptx
Figure 3: Official Rules for IARC
mission8rules.pdfPage updated
Google Sites
Report abuse