Vision/Optical Sensing and Avoidance
Gabriel P. Riccio
7.4 Research Assignment: Sense and Avoid
UNSY 605 Unmanned Systems Sensing, Perception, and Processing
Embry-Riddle Aeronautical University-Worldwide
14Jul16
Small
Unmanned Aerial Systems (sUAS) while being directly piloted by a human have a
zero level of autonomy. The pilot is
responsible for all sensing and avoidance with other aircraft and
obstacles. Semi-autonomous and
autonomous sUAS must have on-board sense and avoidance technologies to reduce
the risk of unwanted collisions. The
Defense Advanced Research Projects Agency (DARPA) has developed a fully
autonomous sUAS quadcopter that uses high definition cameras, LiDAR (Light
Detection and Ranging), sonar, and internal measurement units for sensing and
avoidance (Szondy, 2016). This
combination of sensors has proven very successful for DARPA with their testing
of the quadcopter. However, having
multiple sensors may not be practical for other sUAS applications. Skydio, a new startup company in California,
is working diligently to improve sUAS sensing (Popper, 2015). The company is working to develop
technologies to make sUAS safer and improve autonomous flight capabilities
(Popper, 2015). Skydio engineers believe
they can use standalone ordinary cameras without sonar or lasers to achieve
effective sensing and avoidance (Popper, 2015).
The Phantom 4 semi-autonomous sUAS is equipped with vision sensors for
sensing and avoidance (Bolton, 2016). If
the Phantom 4 flies within 50 feet of an obstacle it will begin to slow, it
comes to a complete stop if it flies within 6 feet of an obstacle (Bolton,
2016). Due to the success of the Phantom
4 and other similar sUAS, vision/optical sensors are an excellent sensor choice
for obstacle avoidance.
Faster and more powerful computers
along with a newer set of algorithms enhance the effectiveness of UAS vision
systems for sUAS (Barry, Oleynikova, Honegger, Pollefeys, & Tedrake, n.d). Vision sensors have proved to be successful
in autonomous flight from takeoff to landing while providing obstacle avoidance
(Barry, et al. n.d.). When sUAS are
outfitted with stereo vision; individual 2-dimensional images are combined to
create 3-dimensional images when appropriately referenced and processed (Barry,
et al. n.d.). Some notable concerns by designers
while selecting vision system for their platform are; latency of the data
stream, power consumption, and the synchronization of multiple image exposures
(Barry, et al. n.d.).
The DJI
Phantom 4 has front obstacle sensors that work in conjunction with its computer
vision and processing to react to and avoid obstacles in its path
("Phantom 4 - DJI’s smartest flying camera ever," 2016). In “Normal Mode” the platform will stop and
hover if an obstacle is in its path, in other modes, it will alter its flight
path to avoid the obstacle or come to a hover if need be ("Phantom 4 -
DJI’s smartest flying camera ever," 2016).
The optical sensing system has a 60 degree by 50 degree field of view
that uses the data collected to create a 3-dimensional map for obstacle
avoidance ("Inside a Drone: Computer Vision," 2016). Additionally, it has dual cameras mounted on
the bottom and dual ultrasonic sensors for position accuracy ("Phantom 4 -
DJI’s smartest flying camera ever," 2016).
The overall weight of the platform is 1380 grams, and has a top speed of
20 meters per second ("Phantom 4 - DJI’s smartest flying camera
ever," 2016). At a retail price of
just under $1400 dollars, it is not cheap, but reasonable with all of its
embedded autonomous capabilities (Popper, 2016). Some important specifications on the Phantom
4 obstacle sensing system based on the DJI company product website are as
follows:
·
Obstacle
Sensory Range – 2 feet to 49 feet
·
Width of
Optical Sensing System – 0.7meters to 15meters
·
Operating
Environment – Surface with clear pattern and adequate lighting (lux>15)
·
Altitude and Operating Range of the Positioning System –
0 feet – 33 feet ("Phantom 4 - DJI’s smartest flying camera ever,"
2016).
In conclusion, the DJI Phantom 4 represents how standalone
vision/optical sensors coupled with fast computing power can be successfully
engineered into sUAS for sensing and obstacle avoidance. If the goal is full autonomy, then the
platform will have to have multiple sensors such as DARPAs’ fully autonomous
quadcopter. The dollar cost for DARPASs’
quadcopter was not presented in any literature during the research for this
paper, however, it can be reasonably hypothesized that DARPA has invested a
considerable amount of money in their project.
For the retail consumer looking for a sUAS with semi-autonomous
functionality, selecting one with vision/optical sensors is an excellent
choice.
References
Barry, A.,
Oleynikova, H., Honegger, D., Pollefeys, M., &
Tedrake, R. (n.d.). Fast onboard stereo for vision UAVs.
Retrieved from http://groups.csail.mit.edu/robotics-center/public_papers/Barry15a.pdf
Bolton, D. (2016, March 2).
DJI unveils the Phantom 4 semi-autonomous drone | News | Lifestyle | The
Independent. Retrieved from http://www.independent.co.uk/life-style/gadgets-and-tech/news/dji-phantom-4-drone-price-buy-autonomous-tapfly-activetrack-a6908096.html
Inside a Drone: Computer Vision.
(2016). Retrieved from http://www.dji.com/newsroom/news/inside-a-drone-computer-vision
Phantom 4 -
DJI’s smartest flying camera ever. (2016). Retrieved from https://www.dji.com/product/phantom-4
Popper, B. (2015, January
15). A tiny startup has made big strides in creating self-navigating drones |
The Verge. Retrieved from http://www.theverge.com/2015/1/15/7550669/skydio-drone-sense-and-avoid-camera-vision
Popper, B.
(2016, March 1). DJI's revolutionary Phantom 4 drone can dodge obstacles and
track humans | The Verge. Retrieved from http://www.theverge.com/2016/3/1/11134130/dji-phantom-4-drone-autonomous-avoidance-tracking-price-video
Szondy, D. (2016, February
12). DARPA's fully-loaded quadcopter autonomously navigates an indoor maze at
45 mph. Retrieved from http://www.gizmag.com/darpa-drone-autonomous-45-mph/41810/
