You are on page 1of 12

Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

Understanding Dangerous Play: A Grounded Theory


Analysis of High-Performance Drone Racing Crashes

Amirreza Barin,1 Igor Dolgov,2 Zachary O. Toups1


1
Play & Interactive Experiences for Learning Lab, Dept. Computer Science
2
Perception, Action, & Cognition in Mediated Artificial & Natural Environments Lab, Dept. Psychology
New Mexico State University, Las Cruces, New Mexico, USA
barin@nmsu.edu, id@nmsu.edu, z@cs.nmsu.edu

A. B. C. D.
Figure 1. A crash in the Drone Racing League’s Level 1 competition (Semifinal 2, Heat 3). The drones are flying over empty seats of a stadium,
approaching a glowing green stand gate. Seen from the first-person view video feed from pilot FlyingBear (yellow), pilot Rafa’s drone (red) hits a
hard-to-spot handrail. In (A.), Rafa’s drone is circled. (B.) is just before the crash. In (C.), Rafa crashes (note the leaderboard shows him eliminated).
In (D.), FlyingBear begins to take action to avoid debris. Images ©2016 DRL; used with permission.

ABSTRACT INTRODUCTION
In drone racing, pilots compete on 3D race courses with quad- Drone racing is a recent competitive sport in which pi-
copter drones. A first-person view camera feed displayed lots, equipped with immersive wearable displays, fly un-
through an immersive head-mounted display, gives them the manned aircraft systems (UAS, i.e., drones) through com-
sense of directly piloting the remote vehicle. The Drone Rac- plex 3D courses. Pilots don a head-mounted display (HMD)
ing League (DRL) operates standardized competitions from wirelessly linked to a front-facing camera mounted on their
which it produces professional video. In the present research, drones, providing them, and their audience, a first-person
we analyze DRL videos to examine the reasons that high- view (FPV) that creates a sense of being on the course (see
performance drone pilots crash. The video corpus, combined video frames in Figure 1). Race courses feature many obsta-
with the standardization of the event, enables us to study pi- cles, requiring pilots to maneuver quickly and carefully. The
lot performance without interfering factors (e.g., customized goal of drone racing is to finish the course in the fastest time
drones or interfaces). We present a deep understanding of the to accumulate points; crashes eliminate pilots from a race.
sport of drone racing, a framework for understanding crashes,
and insights into the reasons that pilots crash, exploring how The sport has proliferated among enthusiasts, attracting am-
the HCI community can benefit drone racing through future ateurs and professionals. There are a small number of pro-
user interface augmentations. fessional drone leagues, including the Drone Racing League
(DRL), which are organized by companies and private in-
vestors. Interest from advertisers and media has been piqued:
Author Keywords
ESPN2, a cable network channel devoted to sports, has regu-
Drone racing; unmanned aircraft systems; UAS; first-person larly featured the DRL since 2016 [34].
video; crashes; grounded theory.
In this paper, we develop grounded theory of drone crashes.
ACM Classification Keywords UAS, in general, experience a much higher rate of crashes
H.5.1 Information Interfaces and Presentation (e.g. HCI): Ar- than their manned counterparts [8, 20]. Documented crashes
tificial, augmented, and virtual realities in drone racing allow scholars to closely investigate causes of
pilot error and better understand drone operation under stan-
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
dardized, stressful conditions. Thus, it is valuable to study
for profit or commercial advantage and that copies bear this notice and the full citation both human performance and interface design in this domain
on the first page. Copyrights for components of this work owned by others than the [9]. While preventing crashes is important for winning races,
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific permission it is vital in non-recreational contexts, where drone crashes
and/or a fee. Request permissions from permissions@acm.org. can cause significant damage to people and property.
CHI PLAY ’17, October 15–18, 2017, Amsterdam, Netherlands
© 2017 Copyright held by the owner/author(s). Publication rights licensed to ACM.
The present research focuses on races conducted in the DRL,
ISBN 978-1-4503-4898-0/17/10. . . $15.00 a professional drone racing league. While other leagues al-
DOI: https://doi.org/10.1145/3116595.3116611

485
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

low racers to provide their drones, DRL rules require racers more crew for commanding the mission and managing pay-
to pilot a standard drone made by the DRL; there is no drone loads, such as sensor packages) [8, 20]. Wwhile drone racers
customization as a component of play. Thus, pilot skill is the pilot via an FPV camera, the majority of UAS are equipped
primary factor in success. This provides a methodological with automation that allows pilots to simply indicate way-
advantage for our research in that we can study drone pilot points on a map and the drone flies on autopilot [2, 4, 5, 32].
performance and crashes using platforms with identical tech- That said, findings gathered from the extant UAS literature
nology running on standardized courses. also apply in the context of drone racing.
We gathered numerous drone racing videos provided by the Unmanned aircraft have incident rates at approximately 10
DRL to identify all crash events. Because these are sporting times greater than their manned counterparts [8, 20]. Al-
events, there are ancillary data sources for crashes (e.g., in- though equipment malfunctions and weather contribute to
terviews with pilots, slow-motion replays, professional com- UAS crashes and mishaps, the primary cause is pilot error
mentary) that help us triangulate the reasons for crashes. In [5,8,20]. Poor pilot performance occurs in various forms and
this paper we present a synthesis of all of these observations often stems from lack of situation awareness [10, 11] and /
to help the community understand the sport and see opportu- or high workload [9]. As with most interfaces, while pilots
nities for design. The research covers one level of DRL racing generally bear the blame, often suboptimal control interface
events with 168 unique videos totaling 514 minutes. We cap- design and / or training are the true source of error [31, 32].
tured a total of 18 documented crashes in the dataset. In this
Despite these challenges, in 2016 the United States Federal
paper, we provide three contributions:
Aviation Administration (FAA) issued the Small Unmanned
1. a first-of-its-kind description for the research community Aircraft Rule (14 CFR Part 107) [45], which allowed routine
of the emerging sport of drone racing; civil operations of small UAS (55 pounds [˜25kg] and less) in
Class G airspace (no more than 400 feet [˜120 meters] above
2. a framework for studying drone crashes; and ground level and not in the vicinity of airports or other desig-
nated areas) [45]. One of the main safety-related stipulations
3. a grounded theory analysis to understand the reasons for to this rule is that UAS pilots must effectively accomplish see-
crashes in drone racing. and-avoid duties, keeping well clear of all other aircraft, per-
It is our expectation that this research will form the founda- sons, and property, by maintaining the aircraft within visual
tion of a better understanding of drone interfaces, drone play, line-of-sight throughout the flight (and occasionally scanning
human performance, and opportunities to design for sports. the sky for incoming air traffic).
At the same time, we present a note of caution about chang- In contrast, DRL pilots are not bound by such rules and fly in
ing the nature of the present sport: while it may be possible to close proximity to each other on closed courses. While typi-
intervene and provide “better” and “safer” drone interfaces, cal UAS pilots look at their aircraft during flight, DRL pilots
they may not be desirable as part of this sport. However, use a fully enclosed FPV HMD, and cannot see their aircraft.
the lessons learned can translate to non-recreational UAS do- Furthermore, while typical UAS pilots rely on assistive tech-
mains where safety is paramount. nologies to navigate, DRL pilots rely entirely on vision.

BACKGROUND UAS Pilot Interfaces


Prior research on drone racing is absent from the academic UAS platforms vary greatly in complexity, and thus, crew
literature, as it is an emerging sport. However, ample atten- composition [2, 4, 5, 8, 20, 46]. Pancock et al. [32] conducted
tion has been paid to other UAS domains in recent decades. a literature review investigating UAS control stations, with
In this section, we briefly cover prior UAS research, elaborate special attention to their designs and capabilities. They found
on the design of UAS pilot interfaces, review visual percep- that that small / simple platforms typically solely require a
tion research, and provide a brief history of drone racing. single external pilot for all phases of flight, whereas big /
complex platforms require an external pilot for takeoff and
Prior UAS Research landing and an internal pilot for other phases of flight.
While drone racers put speed and maneuverability at a pre-
mium, most UAS not designed for this context are designed During manual flight, UAS pilots use common handheld re-
for safety above all else. So, while the standard DRL drone mote controllers, similar to those used with radio-controlled
(RC) aircraft. These devices are typically fitted with two joy-
is bare-bones and consists of a propulsion system, a camera,
sticks and various other functional units, depending on the de-
and a power supply, a standard UAS is equipped with ad-
sign of the specific system [32,48]. On the other hand, for au-
ditional technologies to support tracking (e.g., Global Posi-
tomated flight, external and internal pilots utilize a computer-
tioning System (GPS) sensors), stability (e.g., inertial mea-
suring units, accelerometers), and automation (e.g., sensor- based setup, known as a ground control station. Most ground
driven autopilot) [2, 4, 46]. control stations use a laptop or desktop with mouse and key-
board input, while some include additional input modalities
Additionally, UAS systems vary greatly in terms of platform (e.g., touch screens, joysticks) [32]. The displays include
size and crew composition, often needing an on-site, “exter- some combination of a moving map display, a navigation dis-
nal” pilot for takeoff / landing and a remote, “internal” pilot play, a system health / status display, and an out-window /
responsible for all other phases of flight (in addition to one or payload display [32, 48].

486
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

Visual Perception while Piloting a UAS Recently, research has shown Tau may not be the whole
Dolgov [7] notes that, when manually flying UAS, external story and that individual and stimulus differences impact TTC
pilots use strategies similar to tracking an object on a flat judgment [19]. TTC may be better predicted by image veloc-
projection plane, which is based on a target’s (UAS’s) head- ity, rather than Tau [44]. While more research is needed, the
ing angle, optical trajectory, and optical velocity [30, 37–39]. mastery exhibited by drone racers while using an FPV HMD
Rather than using depth cues, which are mostly absent at to navigate complex 3D courses favors the image velocity ac-
the altitudes and distances at which UAS are operated, pilots count. This is due to the fact that a retinal image originating
use 2D planar cues that enable see-and-avoid duties. When from an HMD does not hold the same relationship to Tau that
UAS platforms are equipped with a forward-facing camera, a bona fide retinal image does. On the other hand, image ve-
UAS pilots can sometimes rely on an FPV out-the-window locity is unaffected by image source.
display to manually navigate the aircraft. In such systems,
this mode of control is typically relied upon during takeoff History of Drone Racing
and landing, whereas the remainder of flight is highly auto- The convergence of a number of technologies (e.g., high-
mated [2, 4, 5, 46]. In contrast, DRL pilots rely entirely on an precision sensors for tracking; miniature cameras; high-
FPV wearable display for all parts of flight. resolution displays; shrinking low-power, high-performance
mechanical components) has driven the proliferation of in-
DRL pilots are adept at executing complex tricks with their expensive and fast drones, leading to the new sport of drone
drones while maneuvering around obstacles at break-neck racing. Drone racing started in late 2014 as an amateur sport
speeds [43]. While this is nothing new in the world of in Australia. Pilots shared video via social media, which
RC aircraft (whose controls are similar to those of external were welcomed by enthusiasts worldwide [49]. As the am-
UAS pilots), DRL pilots accomplish these feats of control ateur play grew in popularity, various organizations helped
while using an FPV wearable display and, effectively, looking cultivate it into a professional sport, resulting in a number of
“through a straw.” While this greatly limits situation aware- competitive leagues and race formats.
ness, DRL pilots can still take advantage of perceptual regu-
larities, like depth and time-to-contact cues, as well as other Professional drone racing is proliferating worldwide. There
information available in their optic flow, to safely and speed- are four large U.S.-based drone leagues which provide com-
ily maneuver their aircraft around the course. petitions with cash prizes: the DRL,1 the MultiGP-Drone
Racing League,2 the Drone Sports Association U.S. National
The concept of optic flow was introduced by Gibson in the Drone Racing Championships,3 and the Aerial Sport League.
1950s and developed over the next several decades [13, 14]. Beyond the U.S., DJI, a Chinese drone maker, has just opened
Optic flow refers to the perceived movement of the optic (vi- a similar arena to the Aerial Sport League’s Drone Sports
sual) array (the totality of what is seen) as person moves World entertainment complex, and the International Drone
through the environment (or the environment moves around Racing Association promotes drone racing worldwide.
them). While DRL pilots are controlling their aircraft re-
motely, they still experience optic flow via their FPV HMD. GROUNDED THEORY METHODOLOGY
Parsimoniously, Gibson originally developed the concept of We aim to understand the sport of drone racing, in general,
optic flow while serving as unit director for the U. S. Army’s and the reasons for crashes, in specific. Grounded theory
Air Force Aviation Psychology Program, where he had access analysis entails collecting a multifaceted corpus of data and
to hoards of pilots. Among numerous feats of human perfor- coding it to derive theory [16, 17]. Similar to other research
mance demonstrated by pilots, one particular scenario that methods, grounded theory requires that researchers construct
piqued his interest was the ease at which ace pilots land their a structured network that identifies the relationships and func-
jets on the deck of an aircraft carrier, a task that seems nearly tionalities of involved participants, data, relevant literature,
impossible given the physics involved. So, Gibson reasoned and collected data [15].
that there must be information available in the optic array that
allows them to achieve such high performance [13, 14]. Grounded theory works well to analyze relationships [6],
which is important when approaching a new field with few es-
Gibson’s thinking was later empirically confirmed by Lee and tablished frameworks. The data we were able to acquire from
others [25, 26, 41]. A property of optic flow is that the clo- the DRL provided multiple, overlapping perspectives: video
sure rate of the gap between an animal and an approach- of drone racing events, with commentary, interviews, and re-
ing object is directly proportional to the rate at which the plays; documentation; and enthusiast websites. Crashes are
retinal projection of the object grows, a variable known as hard to analyze in the wild, but the DRL videos and other doc-
Tau [23, 25, 26, 35, 41]. Tau and its derivative are used by umentation provide multiple perspectives that facilitate trian-
humans and other animals for planning and executing move- gulation, rendering grounded theory an effective approach.
ments and directly perceiving time to contact (TTC) with an
approaching object [26,41]. Accurate perception of Tau is es- In the present research, we used ATLAS.ti Windows4 [3] for
sential for survival of both predatory and prey species, and re- analysis. All videos were downloaded into the software for
quires no complex knowledge. This is best illustrated by the analysis and segmentation.
spectrum of non-human creatures who manage to skillfully 1
https://thedroneracingleague.com/
control their locomotion during complex pursuit, avoidance, 2
http://www.multigp.com/site/index
3
appetitive, and herding behaviors [26, 27, 29, 47]. http://dronesportsassociation.com/
4
http://atlasti.com/product/v8-windows/

487
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

Data Collection and Segmentation Understanding the Sport


In the first step, we collected information about DRL race The overarching goal of the present research was to under-
terms and conditions, piloting rules, venue operation, official stand and communicate to the community the play of drone
course dimensions and boundaries, drone parts, technologies, racing. Through immersing ourselves in the data, we devel-
course timing systems, scoring, pilots, classes, and levels. We oped a deep understanding of drone racing play, pilot and
explored websites that report on the sport (e.g., DroneGuru5 ) spectator experience, and tournament and course design.
and websites that sell drone parts (e.g., AtmosphericAdven-
Investigating Obstacles
tures6 ) to develop a comprehensive understanding.
Through the race videos it became clear that obstacles, which
In the second step, we acquired our primary data source: one are a form of skill test for pilots, are a key part of race course
researcher watched and categorized all available DRL videos. design and worthy of deeper investigation. We turned to
Collection took place over June–October 2016. At that time, the sport documentation, looked specifically at obstacles in
there was a substantial number of videos available from the our videos, and identified the types of obstacles that racers
DRL on YouTube for all races that occurred on DRL Level face. Through our analysis, we also discovered that obstacles
1: Miami Lights (Sun Life Stadium, Miami, Florida). Ad- can be official (constructed) or incidental (i.e., related to the
ditional levels were not available on any network to which venue, but not constructed).
the team had access. At the time of this writing, the DRL is
preparing for a new round of events and is obscuring links to Investigating Crashes
the researched race events. The primary driver for the present research was to investigate
how and why drone racers crash. To advance this compo-
This initial dataset consisted of 284 videos. Some of these nent, we identified crash events in our data, finding 18 events.
were subsets of other videos (e.g., a clip showing only an From these 18 events, we identified all corroborating evi-
interview segment that is part of a larger race video); we re- dence across our dataset and linked them to the crash event.
moved overlapping videos, resulting in a corpus of 168 videos In identifying corroborating documents, the researcher took
totaling 514 minutes. The collected videos were all profes- notes on the event and wrote down transcripts of all speech
sional productions and consisted of four types: associated with the event. On average, there were 4–5 docu-
1. comprehensive race videos that included commentary, ments linked to each of the 18 crashes; the linked documents
multiple event perspectives, FPV videos from all pilots, included interview videos, race videos, FPV videos of the
and interviews (57 videos); crashed drone, other drones’ FPV videos for this crash, notes,
and transcripts.
2. separate pilot interviews (34 videos);
Through the crash dataset, we sought specific information
3. advertisements for DRL events (which included informa- from the data, which emerged as we developed our knowl-
tion on levels beyond the first; 48 videos); and edge base. We reviewed the crash dataset multiple times to
develop our schema. Ultimately, we collected the following
4. instructional videos (29 videos). information for each crash:
The researcher watched all of the videos at least three times • the pilots(s) involved;
each and simultaneously recorded notes. In this phase, he
split videos with multiple contents, mostly the race videos, • the race course and location of the crash on the course;
into their component segments (e.g., race event, post-race in- • the type of obstacle involved (if any); and
terviews, commentary from racers, replays). This process
produced a total of 359 documents (i.e., video segments and • environmental conditions (including ambient light).
unsegmented videos) for further analysis. All these attributes were not always available, but we devel-
In the analysis of the segmented dataset, we categorized oped as much of an understanding as we could.
videos according to content: drone crashes, pilot interviews, Using the data points collected from the linked documents
piloting techniques, course simulation,7 races, and advertise- and coding them, we developed a framing to understand
ments. The categorized videos drove our analysis focus. crashes. In some cases, we found that instructional videos
were helpful in understanding the reasons for a crash. The
Analysis Focus
crash analysis framework synthesizes the information col-
In this research, our aim was to deeply understand the sport of lected and frames each crash in terms of the following fields:
drone racing and identify how and why pilots crash. This fo-
cused our efforts on understanding the mechanics of the sport 1. Crash Location: Locations are on race courses, described
as well as the proximate (e.g., obstacles) and ultimate (e.g., in terms of an obstacle of that course (e.g., a particular tun-
errors in perception) causes of crashes. In the remainder of nel). We have seven crash locations in our dataset.
this section, we explain analysis foci; in the Results section,
2. Crash Position: The exact position in / on an obstacle
we describe what we learned through these foci.
where a drone crashed (e.g., the exact position of the top
5
http://www.droneguru.net of a particular tunnel).
6
http://atmosphericadventures.com/
7
While the other content types are self-explanatory, course simula- 3. Maneuver Type: Maneuver types describe the activity the
tions are video clips that explain the race course, rendered in 3D. pilot was performing with the drone when it crashed (e.g.,

488
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

turning left or right, climbing). Instructional videos were The pilots all fit their HMDs over their faces, securing straps
especially helpful in understanding and describing maneu- and ensuring a comfortable fit. One-by-one they raise one
ver types. hand in a thumbs up, indicating they are set. They are now
fully immersed in the race: their HMD goggles show an FPV
4. Descriptive Reason for Crash: Based on observation video feed from the front cameras of their resting drones.
of race video and interviews, we identify the reasons for They pick up their RC controllers: bulky boxes with two
crashes. thumbsticks and various switches; they prepare to race.
5. Crash Type: We classified crashes as one of three types “Ten seconds, pilots.” A female voice counts down over the
(described in detail later): drone-obstacle collisions, drone- PA system: “Ten, nine, eight...”
drone collisions, or stall crashes. Collisions occur when the
drone strikes another object; stall crashes occur due to the When she reaches “Three”, her voice is replaced by a beep,
aerodynamics of the drone and performed maneuver. another beep, and the starting tone.
This framing is potentially useful to analyze any drone race The drones lift off, one speeding after the other. One lags
crash event. For each data field, there are a number of possi- behind as a commentator notes, “The start can be critical!”
ble values. Some values are specific to our observed venue, The race has begun.
but can be adapted or expanded as new data are collected. A group of three drones bob along at high speed, their pro-
pellers sounding like a humming swarm of angry bees. They
RESULTS swoop over empty stands, cornering around glowing pylons
Our results provide insight into play in the DRL and an under- and through the lit, square gates.
standing of the reasons for drone crashes in races. While our
results are specific to the DRL, mostly due to the extensive The pilots’ “eyes” are the cameras mounted at front of the
number of documents we were able to acquire that were not drones, experiencing the event as if flying extremely fast
available for other leagues or events, we expect that general through a giant’s stadium. Their palms sweat as they impart
understanding can be extrapolated from them. minute changes to the two sticks with their thumbs, tilting
their drones to change their velocity and attitude.8
One contribution is an account of the sport of drone racing
through the lens of the DRL. In this section, we describe The course requires the pilots to deviate from their circular
how the sport is played and provide a detailed description path around the bowl, sweeping out of the stands and over the
of the context of the sport as derived from our observations field. The drones, one-by-one, drop down to where a green
of videos and the sport documentation. The description here illuminated tunnel, normally access for spectators, is the next
combines elements of multiple races that we observed in DRL segment of the course. One drone swings too far, losing time.
Level 1. We describe an example race, from the data, as well They fly through a concrete concessions area, avoiding fly-
as a general description of play. ing too high, lest they hit fans on the ceiling. They dodge
around columns, coming to another tunnel that takes them to
Example Race
a spectator-queuing area. A spiral ramp takes them up a floor
The massive [American] football stadium is devoid of spec- and another tunnel takes them back out into the bowl, where
tators and a course has been configured through it: bright they circle over the stands again.
LED strips line the seats, walls, and columns. The lights
roughly mark out a course through the concrete structure. The final stretch aims the drones at the starting point, where
Aside from pilots’ prior experience flying course, the lights they will encounter the finish box. They pick up speed as
are the only indication that a pilot is on-course. Throughout, they make the final straightaway, at the end of which is a box
large, brightly lit open squares make the gates through which designed to catch the drones and detect who finished first.
drones will fly to score points. One-by-one, the drones smack into the springy back of the
box, falling to its padded floor.
Up to four drones are set up on illuminated podiums at the
starting line. The surface is angled forward, to ensure that
drones take off flying forward, rather than up. They are in Drone Racing League Play
a staggered formation, minimizing the chances of collision. In this section, we provide further details of play. We start
Smoke blows across the podiums, giving the effect that the from our primary data, and generalize to the sport as a whole
brightly lit drones are rockets prepared for takeoff. from less rigorous observations of online sources. Our foci
on understanding the sport and investigating obstacles drive
Nearby, in the safety of a covered VIP area, pilots prepare. our explanations.
They are seated in a row overlooking the stadium in large,
comfortable seats. The vantage point on the empty space is Race Environment
irrelevant, they will be wearing fully immersive HMDs. DRL races take place on 3D courses inside abandoned build-
Radio chatter can be heard, “The course is now closed. We ings, around major landmarks, or inside stadiums. Every
are clearing the area. Initiating check...” DRL event, called a “level” has a unique venue and title (e.g.,
Level 1: Miami Lights, Level 2: L.A.pocalypse). A course
As chatter continues on the radio, an announcer, pacing in the
8
VIP area, says, “Alright, pilots, goggles. Arm check.” Attitude is the angle of the drone relative to horizon.

489
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

that passes through various parts of the venue is identified sible. The box is enclosed, so drones that are stopped (or
and marked out with bright lights and obstacles. disabled from the final impact) are unlikely to escape.
Within the race environment, a safe area for pilots is config- All DRL courses also are designed in with 3D elements. We
ured with a protective cover. Each pilot has a specified seat are able to gather some data on the design of multiple DRL
and may stand up, but they cannot leave the designated pilot courses through simulation videos. We know that they use
area; it is unclear if this is to preserve fairness or for safety. gates, tunnels, ramps, windows, stairs and floors as paths and
obstacles, and that bright LED lighting is key to making these
The drones start the race at a particular location, resting on
things visible to pilots. Each course makes use of the unique
individual podiums that are angled to enable a fast take-off.
affordances of the venue in which they are set.
Additionally, the drones are clustered together, but spaced far
enough apart to ensure that pilots do not collide immediately. We have the most details on the DRL’s Level 1 course, Miami
Lights (Figure 2). This course includes five 3D illuminated
A key piece of preparation is the installation of necessary stand gates. There are also three tunnel segments that create
infrastructure. Infrastructure includes specifying the digital
a tight space for flying. A unique element of this course is the
and analog network channels that are used for communicat-
helix, a long spiraling ramp that rises in elevation.
ing with drones and transmitting FPV video. Cameras are lo-
cated throughout the course to capture the race from multiple Course designs take advantage of drones’ unique ability to
angles for the production of race videos. move in 3D and prioritize the ability to engage in aerial ac-
robatics as a core skill in the game. In any event, there is an
No spectators are allowed within the DRL race environment. identified course that often incorporates 3D elements. Some-
Similar to the DRL, a few other racing leagues utilize indoor
times races are set up through spaces that players cannot
areas, which place hard constraints on drone maneuverability,
physically access, like flying through the windows of aban-
making it probable that some drones will crash. Looking be-
doned buildings. Due to power requirements and cost, only
yond, we find that many organized amateur events are held in
professional leagues are likely to use large amounts of bright
open spaces when/where other people are not present. Racers lights to plot complex pathways.
employ found objects (e.g., trees, playground equipment) as
obstacles or course markers (e.g., to note where pilots should Obstacles
turn around). Play in open spaces can be relatively safe for the DRL courses consist of a number of obstacles, spread
drones, but trees and other obstacles can still be dangerous. throughout a course. Some are official and are configured
as part of the course, others are simply characteristics of the
Live audiences are rare, owing to safety and the fact that
venue (i.e., incidental obstacles).
drones are hard to track visually. This contrasts sharply with
automobile racing, in which crowds can safely and easily The official obstacles in DRL levels are usually gates, tunnels,
watch the large vehicles move around a 2D track [12]. Some- ramps, walls, windows, stairs and floors. Official obstacles in
times, a few spectators watch a race from a safe place, often DRL Level 1: Miami Lights include:
using remote video. The video, typically a direct FPV feed
or transmitted online, is often displayed on the large monitor. • Stand gates are large square or circular frames surrounded
In some races, spectators can choose from the various drone by lights that the drones fly through to score points.
video feeds by tuning their HMDs to the appropriate channel. • Tunnels force drones into tight spaces. These tunnels are
often the passageways that are used to enter or exit the bowl
Courses of the stadium. Sometimes smoke machines spread smoke
Courses are the path drones need to take around a race envi- inside of the tunnels, reducing visibility.
ronment; normally, these are designed to maximize the need
for skillful maneuvering. Drones are free to fly within any • Louvered wall: Pilots must fly along a wall with a repeat-
space in the race environment, but to be competitive they need ing light pattern on it, then break away at the right time.
to pick efficient lines in navigating the course. For any DRL • Helix: A winding spiral ramp up which pilots must fly.
race, the course is run by pilots one time, accumulating points
by clearing obstacles and finishing quickly. In addition, the seats, handrails, ceiling fans, and walls of the
stadium serve as incidental obstacles that pilots must avoid.
The starting line for a course includes the podiums for the
drones. Each course consists of a number of obstacles, pri- For general drone racing, found objects often serve as obsta-
marily gates, that drones must pass through. The courses are cles to avoid. Inasmuch as the goal of the sport is to demon-
lit throughout; in some cases, the lights are the only marker strate acrobatic ability, obstacles serve to engage pilots in dis-
of where the course actually is. For example, part of Level plays of skill.
1 involves the drones flying out across an open space toward
Equipment
the football field in the middle; a line of lights marks where
they need to drop down and turn around to enter a tunnel. For DRL races, all equipment is standardized, so equipment
customization (including drones) is not a component of play.
Each course concludes with a finish box: a large, padded box DRL funds a special engineering department for developing
with one open side to catch speeding drones. The design of drones, video transmission methods, radio frequency technol-
the box is such that it allows drones to crash into its interior ogy, and race equipment. For each pilot, normal DRL equip-
somewhat safely, supporting finishing the race as fast as pos- ment includes a drone, RC controller, and HMD.

490
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

Controller. RC controllers are used to pilot drones and con-


sist of two analog thumbsticks that can be pushed in any di-
rection: the right joystick controls roll (via left-right move-
ment) and pitch (via forward-back movement) and the left
joystick controls yaw (via left-right movement) and overall
throttle (via forward-back movement). Roll is the primary
means to turn a drone while flying, causing the drone to tilt
along its longitudinal axis by adjusting the relative power to
rotors on either side. Pitch adjusts the forward / backward
leaning of the drone, which enables it to fly forward or back-
ward. Yaw is primarily useful for spinning the drone in place.
As with all rotor-wing aircraft, the throttle controls how fast
all props are spinning, so that, if tilted forward, the drone
acquires forward and upward momentum; if level, the drone
will simply go up.
This is a standard configuration for drone racers (and RC
flight, in general). The overall effect is that, to fly forward,
the pilot pushes both sticks forward. Typical turns are accom-
plished by moving the left stick to the left or right, causing the
drone to bank.
Figure 2. Overview of the DRL Level 1 Miami Lights course in Sun Life
Stadium. A blue flag calls out the starting point and the course follows The controller and drone must be set to the same radio fre-
the bright blue line around the bowl of the stadium, into the side areas, quency channels, which are specified for each pilot in a race.
along the louvered wall (bottom center), via the three tunnels (one near
Checkpoint A, lower-left; two near Checkpoint B, lower-right), up the HMD Goggles. DRL pilots wear HMD goggles that com-
helix (lower-right of image), around the bowl again, and concludes with pletely obscure vision of the world around them. The HMD
the finish box at the black Finish Line flag. Image ©2016 DRL; used goggles provide a raw live video feed of the drone’s FPV
with permission. camera, simulating sitting in a cockpit.
In the DRL, race status information is overlaid on the live
video feed to each racer. The overlay provides a leaderboard
in the corner, which indicates each pilot’s relative position
Drone. Drones are the primary technology for the DRL and whether or not any have crashed. This is accomplished
and the focus of its development efforts, which optimize for with special software and a repeater device.
speed, agility, and resilience. In the present research, pilots
Infrastructure
were using Racer 2 drones, although DRL has recently re-
A key part of any race is setting up infrastructure. All course
leased the Racer 3. DRL provides all drones for its competi-
designers must be careful about ensuring that the radio bands
tions and keeps a queue of ready drones for each pilot as races
needed for transmitting to and from drones are interference
progress.
free. In addition, the DRL includes substantial infrastructure
Racing drones are quadcopters (4 upward facing rotors), but for live mixing of video feeds and for recording events.
are configured differently from common hobby drones. Rac-
ing drone rotors form an H-shape, rather than an X-shape, to Race Scoring and Structure
support forward speed, rather than stability. DRL drones use DRL races consist of heats9 with up to four pilots in each.
powerful motors and aim to be lightweight. Since they impact Each pilot will compete in multiple heats, accumulating
the finish box and crash land, in addition to normal crashes, points for a race; the winner of the race gets the most points.
they must have a sturdy frame and body. A crash eliminates a pilot from a particular heat, but that pilot
can still earn points by competing in other heats. Points are
The front-mounted camera is angled such that pilots can see scored as follows:
straight ahead as the drone pitches forward during flight. The
camera transmits video in real time to the pilot’s HMD; the For passing at least two checkpoints and visiting the
feed can be tapped into for spectating purposes. course in its entirety, pilots are awarded 50 points. For
every second underneath the two-minute time cap, pilots
As they are competing in races, visually tracking drones is are awarded an additional 10 points. Each race features
key, both for spectators and pilots. DRL drones are covered a number of heats. [The final point total is calculated by
in all sides in a grid of color-changable LEDs. During each summing the points across all the heats]. [49]
race, a pilot’s drone is set to a particular color.
Race Regulations
Outside the DRL, drone customization is often a key com- Leagues specify their own regulations for competition, like
ponent of competition and a means for pilots to distinguish any sport. These cover the following:
themselves. Racing drones run a gamut of designs, but are
9
almost universally small quadcopters with an FPV camera. A heat is a single competitive run through a course.

491
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

• Aircraft specifications for frame shape and size, propeller • drone-obstacle collisions occur when a drone strikes an
shape and size, motors, batteries, and vehicle weight. obstacle, either official or incidental;

• FPV camera and goggles may be subject to requirements • drone-drone collisions occur when two racers acciden-
about what data can be shown. tally collide mid-flight; and

• Safety rules cover how people act within a venue, in par- • stall crashes occur when a racer places the drone at an
ticular staff and pilots. They also cover rules about pre- attitude such that it has insufficient lift to stay airborne.
flight inspections, allowed types of flight, first aid, and fire In our dataset, we observed 15 drone-obstacle collisions, 1
hazards. One DRL safety rule is that the first step in re- drone-drone collision, and 2 stall crashes.
trieving any crashed drone is to pull the power connector
off to ensure its rotors do not accidentally activate. Collisions
Understanding the reason for collisions (both drone-obstacle
• Participant rules and requirements cover how league and drone-drone) was a key part of this research. We were
personnel and spectators should act. able to triangulate crash reasons by looking at multiple linked
documents for each crash event. In many cases, racers who
• Judgment rules identify how ambiguous situations in a crashed were interviewed to understand their reasoning. In
race will be adjudicated. addition, since crashes were a focus of the produced race
videos, we were often able to observe each crash from the
• Frequency management rules describe requirements for racer’s viewpoint, the viewpoint of trailing racers, and the
radio transmission within the venue to ensure all pilots viewpoint of venue cameras.
maintain a video feed and control without interference.
Regarding the one drone-drone collision, while we take this
• Race levels and classes identify the number and type of as its own type, these types of collision are rare. The one
levels in a competition. instance observed involved two drones taking off simultane-
ously and out of sight of one another.
• Scoring structure specifies how pilots accumulate points
and win races. Stall Crashes
While the mechanics of collisions are straightforward, stall
Tournament Structure crashes may be unfamiliar to the reader. Racing drones in the
Competition in the DRL starts with the pre-season, then DRL are quadcopter designs with four rotors that provide lift
moves through each of the levels, which culminate in Level in one direction [33, 42]. If the pilot changes the attitude of
5. The competition begins with 16 pilots total, four of whom the drone and exceeds its critical angle of attack, the drone
are eliminated in the pre-season event. The remaining 12 pi- may slow below its stall speed in level flight, which renders it
lots compete to win each of the five levels, based on points uncontrollable.
accumulated through a number of heats on each of those
An example of a stall crash was when pilot Hazak crashed in
courses. The heats are organized into races, creating a tour-
Qualifying Race 2, Heat 1. He was banking over the (empty)
nament structure within the level. Points and eliminations do
stands through a series of gates. After clearing the last gate,
not carry forward to the next level.
he banked especially hard, pushing the drone to be perpendic-
In general, racing events have qualifying levels that they hold ular to the ground. When that happened, his drone only had
in either a regional or non-regional format. A normal struc- thrust to its left and none upward, gravity caused it to drop
ture is to have a tournament with semifinals, finals, and a out of the race and into the stands.
championship to eliminate players and declare a champion.
Proper drone piloting techniques prevent stall crashes, as we
learned through instructional videos. Avoiding stalls is a key
Crash Analysis skill that is learned through practice.
A major focus of the present research was to explicate the
reasons for drone crashes, and consider their role in the sport. Crash Reasons
Crashes in the DRL result in pilots being eliminated from the In interviews, pilots discussed the reasons for crashes in terms
current heat, but not the race, which is a substantial penalty. of human performance. While they mentioned that stress and
At the same time, in other competitions, the stakes can be anxiety are factors for crashes, due to the high stakes of the
higher (e.g., loss of an expensive drone, complete loss of a competition, many pointed to a failure of maneuvering skills
competition). We thus expect understanding such incidents as the main reason for crashing. They pointed to two issues:
to be vital. choosing the wrong maneuver at the wrong time and a failure
to execute the maneuver correctly.
An interesting point in the DRL is that all drone hard-
ware is standardized and racers practice with that hardware. Our study shows that collisions occurred due to three main
This helps to eliminate any problems arising from unfamiliar reasons: improper maneuvering, errors in perception, and in-
equipment that could lead to crashes. terference from lighting transitions.
From our analysis of crash events, we identified three possible Improper Maneuvering. While racing drones are agile, they
crash types: also move very fast. Many collision crashes were a result of

492
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

the failure of the pilot to maneuver through a tight space. We As in car racing, the possibility of crashes increases the
identified improper maneuvering as the reason for 7 of the drama of races and adds excitement. There is an open ques-
drone-obstacle collisions in different places as well as the two tion as to whether or not reducing crash rate in professional
stall crashes. drone races is desirable, although in other contexts (e.g., non-
entertainment flight; drone races with privately owned, ex-
For example, at one of the sections of Level 1, after drones pensive drones) it clearly is.
exit from a tunnel they must turn right to the helix immedi-
ately. This transition resulted in many of the drone-obstacle None of the pilot interview data suggested a need for tech-
collisions due to the degree of the turn and the speed at which nological intervention; in fact, using a bare-bones design is
pilots were flying. For instance, pilot M0ke crashed in Final what enables a DRL drone’s speed and agility. The consen-
Heat 3 upon his drone exiting from a tunnel and wanting to sus among pilots indicates that they value the opportunity to
turn right into the helix. Unfortunately for M0ke, he immedi- engage in complex maneuvering and enjoy flying fast. Fur-
ately collided with a column. ther, crashes are simply a component of the sport.
Errors in Perception. Errors in visual perception were the
second-most prevalent reason for drone racing crashes in our
dataset. While racing, pilots must use TTC, depth cues, and
other information in their optic flow to choose the lines that Potential for User Interface Improvements
they will fly through a course. However, due to the limited While crashes seem integral to drone racing, we have two
field of view of the front-mounted camera, pilots’ situation perspectives on how more advanced UIs could improve the
awareness is compromised. Moreover, human perception is sport while preserving the value of crashes.
imperfect at extreme speeds, thus racers have difficulty judg- First, DRL pilots represent some of the best drone racing pi-
ing their potential to avoid obstacles, causing collisions. Er- lots in the world, they are extremely skilled and have prac-
rors in perception were the reason for 6 of the observed drone- ticed their craft extensively. The process of undertaking this
obstacle collisions. practice is time-consuming and expensive, and so augmented
For instance, when pilot Hazak in the DRL race Semifinal 2 drone interfaces could serve to speed up pilot education and
Heat 3 was preparing to enter a tunnel gate, he lined up his better prepare them for DRL-style drone racing.
drone and flew straight for a full 3 seconds. Believing his Second, although skilled, pilots are playing through a techno-
drone was aligned with the entrance, he flew directly into the logical mediation that does not support human cognition ef-
top of the tunnel. fectively. This failure, especially errors in perception, appears
Interference from Lighting Transitions. to directly translate into the few crashes that we observed.
Further, unlike pilots of manned aircraft, drone pilots lack
Some crashes were caused by drones moving from a dark the physical experience that lends them information about the
space into a bright one or vice-versa. This lighting transi- aerodynamic state of the aircraft. They also fly without basic
tion, which lasts only a moment, makes it hard for the racer instruments that are normally needed for flight.
to identify and avoid obstacles, leading to collisions. This
was the reason for 2 drone-obstacle collisions. Prior research in the community has focused on how drones
might be used as a source of design, but not focused on how
to support piloting them. A number of projects have con-
DISCUSSION sidered using drones for display purposes (e.g., [18, 22, 36]),
Drone racing user interfaces are simple and direct, modeling for playing games (e.g., [22, 28]), or for remote collaboration
the basics of flight. A mental model of quadcopter flying is (e.g., [21]). The UI to control racing drones is, presently, sim-
needed, which is likely readily acquired through practice, in ple, and only a few projects have considered how new tech-
order to manipulate a drone. Further practice helps pilots de- nologies might support more interesting control devices and
velop an intuitive sense of the aircraft’s aerodynamic capabil- control mappings (e.g., [1, 24]). The projects fall short of ad-
ities and limitations, especially with regards to stalls. Based dressing the issues we have discovered in drone racing.
on our observed data, an understanding of how the sport is
played, an understanding of the practice of spectating, and an Changes to drone UIs could help to develop human skill for
understanding of crashes, we expect that there is ample op- more exciting races or could directly improve performance.
portunity for new UI designs in drone racing. In the remainder of this section, we look at the ways in which
the research community could improve drone UIs for the
purposes of racing. Beyond this scope, these improvements
Value of Crashes
could benefit drone pilots, especially training, in general.
Watching pilots’ performance is a primary motivator for spec-
tators in drone racing and a primary concern in the design of While we expect that faster processing, better display tech-
the DRL. The controlled environment of the DRL helps to nologies, and faster drones will enhance the appeal of the
ensure that pilot skill is the primary point of competition (and sport, these advancements are straightforward and largely di-
the reason for crashes). Careful attention is paid to ensuring rected at hardware. At the same time, such hardware im-
that flying drones are consistent, that pilots’ capabilities are provements are likely to exacerbate the existing challenges
not technologically hindered, and that everyone has a fair and of flight, pushing it beyond human capabilities. This argues,
equivalent experience. further, for the need for UI enhancements to the sport.

493
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

Supporting Accurate Perception An alternative solution is to have spectators located safely in


One part of our results shows that most crashes are due to an area where they can enjoy the race live through HMD gog-
collisions, and the main reasons for collisions are failures in gles. An augmented view of the race could highlight racers
human perception and performance. This is illustrated by the and provide insight into their movement vectors. Spectators
fact that many crashes happened when racers tried to enter a could select their preferred video feed and engage with the
check point or clear an obstacle but erred in their perception race at a greater depth.
or maneuvering.
CONCLUSION
UI support could be added to support diagnosis of passabil- In this paper, we have performed a grounded theory analysis
ity through and around various obstacles. Depending on the of a large corpus of DRL video data to deeply understand the
stance of drone racers, this could function in a tutorial mode sport of drone racing and develop an understanding of the rea-
to enable new pilots to understand the inherent difficulties in sons for crashes. We contribute a first-of-its-kind description
judging distance through FPV HMD goggles, yet we expect of the sport for consumption by the research community. The
it to be valuable even as a prosthesis to overcome the inherent research has led us to a framework for describing and charac-
deficiencies in the setup. terizing crashes. Looking further, we point toward the value
We envision simple augmentations to a pilot’s video feed: in- of interfaces in this sport and the benefits they could provide.
dicators on the feed of where the drone is projected to be in It is our expectation that the sport of drone racing will con-
the near future. More advanced designs could warn of the tinue to advance, pushing the limits of human capability and
location of an impending crash. that, as a sport that is inherently technologically mediated,
Technologically, supporting accurate perception can likely be both for participants and spectators, it stands to benefit from
performed through computer vision analysis of live video, to leveraging that mediation. Looking to the future of drone rac-
minimize the impact on the lean design of drones. At the ing, we expect that improving pilots’ and spectators’ inter-
same time, augmenting drones with a combination of sensors faces are the keys to enabling more exciting and entertaining
and / or data on the 3D configuration of their race environ- races. Such developments require that drone platforms ad-
ment would enable similar capabilities. vance, along with the sophistication of their sensor packages
and ability to convey data to the pilot and transmit nuanced
Supporting Instrumentation control to the drone.
Stall crashes occur when a drone’s angle of attack is such that
it loses lift. Stalls are a known issue in flight (both in tradi- We close with a quote from Stephenson’s Snowcrash, in
tional contexts and drone racing), and we observed a number which he describes characters engaged in racing in a virtual
of training videos and resources aimed at supporting drone world (the Metaverse). The description bears a striking re-
racers in avoiding stalls. Drone pilots have to be made aware semblance to the experience of drone racing. For us, this
of the drone aerodynamic states that can lead to stall crashes quote highlights how drone racing can advance as a sport, but
during maneuvers. Drone racers lack even the most basic in- one in which UI advancements are vital in supporting physi-
struments and this, combined with being outside the cockpit cal advances in racing performance.
of their vehicles, increases the likelihood of stall crashes. A Metaverse vehicle can be as fast and nimble as a
We thus suggest that even a basic stall indicator would sup- quark. There’s no physics to worry about, no constraints
port drone pilots immensely. Such as system could commu- on acceleration, no air resistance. Tires never squeal and
nicate, through iconography, the lift state of a drone by using brakes never lock up. The one thing that can’t be helped
small, onboard sensors. Again, even used as a training tool, is the reaction time of the user. So when they were racing
such an indicator could help drone pilots avoid stall crashes. their latest motorcycle software through Downtown at
Mach 1, they didn’t worry about engine capacity. They
Enhancing the Spectator Experience worried about the user interface, the controls that en-
Drone racing is primarily watched in a technologically medi- abled the driver to steer, accelerate, or brake as quickly
ated way (e.g., professionally produced video from multiple as he could think. Because when you’re in a pack of
viewpoints, large live screens, HMDs that can tune to mul- bike racers going through a crowded area at that speed,
tiple drones’ frequencies). Spectators are rarely near a race and you run into something and suddenly slow down to
because it is unsafe. Further, the inability to spot speeding a speed of exactly zero, you can forget about catching
drones with the naked eye is challenging even at a close dis- up. One mistake and you’ve lost. [40, p. 354]
tance.
ACKNOWLEDGEMENTS
A straightforward solution for visibility is to add lighting the This material is based upon work supported by the National
drones. Some leagues provide or require drones that have Science Foundation under Grant No. IIS-1619273.
bright LEDs on them, as is true of the DRL’s Racer drones.
Races are held indoors or at night for this reason.

494
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

REFERENCES 12. Mattias Esbjörnsson, Barry Brown, Oskar Juhlin, Daniel


1. Charlie Anderson, Benji Barash, Charlie McNeill, Denis Normark, Mattias Östergren, and Eric Laurier. 2006.
Ogun, Michael Wray, Jarrod Knibbe, Christopher H. Watching the Cars Go Round and Round: Designing for
Morris, and Sue Ann Seah. 2015. The Cage: Towards a Active Spectating. In Proceedings of the SIGCHI
6-DoF Remote Control with Force Feedback for UAV Conference on Human Factors in Computing Systems
Interaction. In Proceedings of the 33rd Annual ACM (CHI ’06). ACM, New York, NY, USA, 1221–1224.
Conference Extended Abstracts on Human Factors in DOI:http://dx.doi.org/10.1145/1124772.1124955
Computing Systems (CHI EA ’15). ACM, New York,
13. James J. Gibson. 1950. The Perception of the Visual
NY, USA, 1687–1692. DOI:
World. The Riverside Press, Cambridge, Massachussets,
http://dx.doi.org/10.1145/2702613.2732877
USA.
2. Ella Atkins, Anı́bal Ollero, and Antonios Tsourdos
14. James J. Gibson. 2014. The Ecological Approach to
(Eds.). 2016. Unmanned Aircraft Systems. John Wiley &
Visual Perception (classic ed.). Psychology Press, New
Sons Ltd., Chichester, West Sussex, United Kingdom.
York, New York, USA.
3. ATLAS.ti Scientific Software Development GmbH.
2017. ATLAS.ti Windows version 8. Software 15. Barney G. Glaser. 1978. Theoretical Sensitivity:
[Windows]. (2017). Advances in the Methodology of Grounded Theory. The
http://atlasti.com/product/v8-windows/
Sociology Press, Mill Valley, California, USA.

4. Reg Austin. 2010. Unmanned aircraft systems: UAVS 16. Barney G. Glaser. 1998. Doing Grounded Theory:
design, development and deployment. John Wiley & Issues and Discussions. The Sociology Press, Mill
Sons Ltd., Chichester, West Sussex, United Kingdom. Valley, California, USA.

5. HaiYang Chao, YongCan Cao, and YangQuan Chen. 17. Barney G. Glaser and Anselm Strauss. 1967. The
2010. Autopilots for small unmanned aerial vehicles: a Discovery of Grounded Theory: Strategies for
survey. International Journal of Control, Automation Qualitative Research. Aldine Publishing Company,
and Systems 8, 1 (2010), 36–44. Chicago, Illinois, USA.

6. Juliet Corbin and Anselm Strauss. 2008. Basics of 18. Antonio Gomes, Calvin Rubens, Sean Braley, and Roel
Qualitative Research: Techniques and Procedures for Vertegaal. 2016. BitDrones: Towards Using 3D
Developing Grounded Theory (3rd ed.). SAGE Nanocopter Displays As Interactive Self-Levitating
Publications, Inc, Thousand Oaks, California, USA. Programmable Matter. In Proceedings of the 2016 CHI
Conference on Human Factors in Computing Systems
7. Igor Dolgov. 2016. Moving towards Unmanned Aircraft (CHI ’16). ACM, New York, NY, USA, 770–780. DOI:
Systems Integration into the National Airspace System: http://dx.doi.org/10.1145/2858036.2858519
Evaluating Visual Observers’ Imminent Collision
Anticipation during Day, Dusk, and Night sUAS 19. Peter A. Hancock and M. P. Manster. 1997.
Operations. International Journal of Aviation Sciences Time-to-Contact: More than Tau Alone. Ecological
1, 1 (2016), 41–56. Psychology 9, 4 (1997), 265–297. DOI:
http://dx.doi.org/10.1207/s15326969eco0904_2
8. Igor Dolgov and Stephen B. Hottman. 2011. Human
Factors in Unmanned Aircraft Systems. In Introduction 20. Alan Hobbs. 2010. Unmanned Aircraft Systems. In
to Unmanned Aircraft Systems, Richard K. Barnhart, Human Factors in Aviation (2nd ed.), Eduardo Salas and
Stephen B. Hottman, Douglas M. Marshall, and Eric D. Maurino (Eds.). New York: Elsevier, 505–531. DOI:
Shappee (Eds.). CRC Press, 165–180. DOI: http://dx.doi.org/10.1016/B978-0-12-374518-7.
http://dx.doi.org/10.1201/b11202-12 00016-X

9. Igor Dolgov, Elizabeth Kaltenbach, Ahmed S. Khalaf, 21. Brennan Jones, Kody Dillman, Richard Tang, Anthony
and Zachary O. Toups. 2017. Measuring human Tang, Ehud Sharlin, Lora Oehlberg, Carman
performance in the field: Concepts and applications for Neustaedter, and Scott Bateman. 2016. Elevating
unmanned aircraft systems. In Human Factors in Communication, Collaboration, and Shared Experiences
Practice: Concepts and Applications, H. Cuevas, in Mobile Video Through Drones. In Proceedings of the
J. Velasquez, and A. Dattel (Eds.). Taylor & Francis. In 2016 ACM Conference on Designing Interactive Systems
press. (DIS ’16). ACM, New York, NY, USA, 1123–1135.
DOI:http://dx.doi.org/10.1145/2901790.2901847
10. Mica R. Endsley. 1995. Toward a theory of situation
awareness in dynamic systems. Human Factors 37, 1 22. Matjaž Kljun, Klen Čopič Pucihar, Mark Lochrie, and
(1995), 32–64. Paul Egglestone. 2015. StreetGamez: A Moving
Projector Platform for Projected Street Games. In
11. Mica R. Endsley. 2000. Theoretical Underpinnings of Proceedings of the 2015 Annual Symposium on
Situation Awareness: A Critical Review. In Situation Computer-Human Interaction in Play (CHI PLAY ’15).
Awareness Analysis and Measurement, Mica R. Endsley ACM, New York, NY, USA, 589–594. DOI:
and D. J. Garland (Eds.). Lawrence Erlbaum Associates, http://dx.doi.org/10.1145/2793107.2810305
Mahwah, NJ, USA, 3–6.

495
Session 8: Beyond the Typical Player Experience CHI PLAY 2017, October 15–18, 2017, Amsterdam, NL

23. Jan J. Koenderink. 1986. Optic flow. Vision Research 26, 36. Stefan Schneegass, Florian Alt, Jürgen Scheible,
1 (1986), 161–179. Albrecht Schmidt, and Haifeng Su. 2014. Midair
Displays: Exploring the Concept of Free-floating Public
24. Nataliya Kos’myna, Franck Tarpin-Bernard, and
Displays. In CHI ’14 Extended Abstracts on Human
Bertrand Rivet. 2014. Bidirectional Feedback in Motor Factors in Computing Systems (CHI EA ’14). ACM,
Imagery BCIs: Learn to Control a Drone Within 5 New York, NY, USA, 2035–2040. DOI:
Minutes. In CHI ’14 Extended Abstracts on Human http://dx.doi.org/10.1145/2559206.2581190
Factors in Computing Systems (CHI EA ’14). ACM,
New York, NY, USA, 479–482. DOI: 37. Dennis M. Shaffer, Igor Dolgov, Andrew Maynor, and
http://dx.doi.org/10.1145/2559206.2574820 Cody Reed. 2013a. Receivers in American football use a
constant optical projection plane angle to pursue and
25. David N. Lee. 1976. A theory of visual control of catch thrown footballs. Perception 42, 8 (2013),
braking based on information about time-to-collision. 813–827.
Perception 5, 4 (1976), 437–459.
38. Dennis M. Shaffer, Igor Dolgov, Eric Mcmanama,
26. David N. Lee and Paul E. Reddish. 1981. Plummeting Charles Swank, Andrew B. Maynor, Kahlin Kelly, and
gannets: A paradigm of ecological optics. Nature 293 John G. Neuhoff. 2013b. Blind (fold) ed by science: A
(1981), 293–294. DOI: constant target-heading angle is used in visual and
http://dx.doi.org/10.1038/293293a0 nonvisual pursuit. Psychonomic bulletin & review 20, 5
27. David N. Lee, Paul E. Reddish, and D. T. Rand. 1991. (2013), 923–934.
Aerial docking by hummingbirds. Naturwissenschaften 39. Dennis M. Shaffer, Richard S. Marken, Igor Dolgov,
78, 11 (1991), 526–527. and Andrew B. Maynor. 2013c. Chasin’ choppers: using
28. Joseph Lindley and Paul Coulton. 2015. Game of unpredictable trajectories to test theories of object
Drones. In Proceedings of the 2015 Annual Symposium interception. Attention, Perception, & Psychophysics 75,
on Computer-Human Interaction in Play (CHI PLAY 7 (2013), 1496–1506.
’15). ACM, New York, NY, USA, 613–618. DOI: 40. Neal Stephenson. 1992. Snow Crash. Bantam Books,
http://dx.doi.org/10.1145/2793107.2810300 New York, NY, USA.
29. Jack M. Loomis and Andrew C. Beall. 1998. Visually 41. H-J Sun, DP Carey, and MA Goodale. 1992. A
controlled locomotion: Its dependence on optic flow, mammalian model of optic-flow utilization in the
three-dimensional space perception, and cognition. control of locomotion. Experimental Brain Research 91,
Ecological Psychology 10, 3-4 (1998), 271–285. 1 (1992), 171–175.
30. Michael K. McBeath, Dennis M. Shaffer, and Mary K. 42. The Drone Racing League. 2016a. Proximity Stall.
Kaiser. 1995. How baseball outfielders determine where (2016). https://thedroneracingleague.com/
to run to catch fly balls. Science 268, 5210 (1995), 569. trick-wiki/proximity-stall/.
31. Don Norman. 2013. The Design of Everyday Things: 43. The Drone Racing League. 2016b. Trick Wiki. (2016).
Revised and Expanded Edition (2013 ed.). Basic Books, https://thedroneracingleague.com/trick-wiki/.
New York, New York, USA.
44. James R. Tresilian. 1999. Visually timed action:
32. Carl Pancock, Jr., Ellen J. Bass, and Philip J. Smith. Time-out for ‘tau’? Trends in Cognitive Sciences 3, 8
2016. UAS human factors and control station design (1999), 301–310.
standards: UAS control station literature review.
Technical Report. 45. U.S. Government Publishing Office. 2017. Small
Unmanned Aircraft Systems, 14 C.F.R. Part 107. (2017).
33. Quadcopter Flight School. 2014. Vortex Ring State (The
Wobble of Death). (2014). 46. Kimon P. Valavanis, P. Oh, and Les A. Piegl. 2013.
http://quadcopter101.blogspot.com/2014/10/ Unmanned aircraft systems. Springer.
vortex-ring-state-wobble-of-death.html. 47. Hermann Wagner. 1982. Flow-field variables trigger
34. Darren Rovell. 2016. Drone Racing League, ESPN landing in flies. Nature 297 (1982), 147–148.
reach broadcasting agreement. (2016). http: 48. Adam C. Watts, Vincent G. Ambrosia, and Everett A.
//www.espn.com/moresports/story/_/id/17544727. Hinkley. 2012. Unmanned aircraft systems in remote
35. G. J. Savelsbergh, H. T. Whiting, and Reinoud J. sensing and scientific research: Classification and
Bootsma. 1991. Grasping Tau. Journal of Experimental considerations of use. Remote Sensing 4, 6 (2012),
Psychology: Human Perception and Performance 17, 2 1671–1692.
(1991), 315. 49. Jesse Young. 2017. Everything You Need to Know about
Drone Racing. (2017). http://www.droneguru.net/
everything-you-need-to-know-about-drone-racing/.

496

You might also like