Use of a Small Unmanned Aerial System for the SR-530 Mudslide Incident near Oso, Washington

The Center for Robot-Assisted Search and Rescue deployed three commercially available small unmanned aerial systems (SUASs)—an AirRobot AR100B quadrotor, an Insitu Scan Eagle, and a PrecisionHawk Lancaster—to the 2014 SR-530 Washington State mudslides. The purpose of the flights was to allow geologists and hydrologists to assess the eminent risk of loss of life to responders from further slides and flooding, as well as to gain a more comprehensive understanding of the event. The AirRobot AR100B in conjunction with PrecisionHawk postprocessing software created two-dimensional (2D) and 3D reconstructions of the inaccessible “moonscape” region of the slide and provided engineers with a real-time remote presence assessment of river mitigation activities. The AirRobot was able to cover 30– 40 acres from an altitude of 42 m (140 ft) in 48 min of flight time and generate interactive 3D reconstructions in 3 h on a laptop in the field. The deployment is the 17th known use of SUAS for disasters, and it illustrates the evolution of SUASs from tactical data collection platforms to strategic data-to-decision systems. It was the first known instance in the United States in which an airspace deconfliction plan allowed a UAS to operate with manned vehicles in the same airspace during a disaster. It also describes how public concerns over SUAS safety and privacy led to the cancellation of initial flights. The deployment provides lessons on operational considerations imposed by the terrain, trees, power lines, and accessibility, and a safe human:robot ratio. The article identifies open research questions in computer vision, mission planning, and data archiving, curation, and mining. 1 digitalcommons.unl.edu 2 M u r p h y e t a l . i n J o u r n a l o f F i e l d R o b o t i c s ( 2 0 1 6 ) protect responders working on recovering victims and mitigating flooding. Figure 1 shows the slide. The riparian section of the slide, dubbed the “moonscape,” and the intersection of the scarp and moonscape, called the “toe,” were inaccessible by foot or ground vehicle due to quicksand-like mud that was over 6 meters deep. The moonscape, toe, and lower portions of the scarp could not be sensed with satellite remote sensing with sufficient resolution. Manned helicopters could not acquire a complete survey because of the need to stay at altitudes higher than 500 feet for safety reasons, as the narrow canyon produces unpredictable gusts and there was a danger that debris loosened by the rotor wash would be sucked into the rotors and cause a crash. The expense of manned helicopters precluded daily use for a rapidly changing situation: the rule of thumb from the 2014 AUVSI/AIAA Civilian Applications of Unmanned Aerial System (CAUAS) workshop was that manned helicopters cost 10 times more than a SUAS, i.e., about $250 an hour versus $25 an hour. The available LIDAR data from manned assets were taking 2–4 days for processing and subsequent release to the responders in the field. The Texas A&M Engineering Experiment Station’s Center for Robot-Assisted Search and Rescue (CRASAR) provided three commercially available small SUASs—two fixed-wings and one rotorcraft—through its Roboticists Without Borders (RWB) program. The flights make at least three contributions. First, they were the first SUAS flights reported specifically for mudslides. Thus they add another case study to the growing corpus of SUAS applications, as well as providing insight into platforms, operations, and open research questions. Second, the flights exemplify the evolution of SUASs from data collection platforms to data-to-decision systems, where the system collects data and converts it to actionable information readily comprehended by decision makers. Third, the flights illustrate the increasing impact of regulations and societal concerns. SUAS flights for the 2013 floods in Boulder, CO were suspended due to the lack of adherence to regulations (9News, 2013). The SR-530 flights reported in this article occurred under a novel airspace deconfliction plan approved by the Federal Aviation Agency that allowed manned operations in the same area. However, misunderstandings about regulations within the emergency response community and public perception of privacy caused initial flights to be cancelled. The article is organized as follows. Section 2 reviews previous and related work in disaster robotics. Section 3 describes the general missions and selection of platforms using the criteria established in Murphy (2014). Section 4 describes the initial deployment in March 2014, which resulted in zero flights. The Insitu Scan Eagle could not find a staging area that had sufficient space for launch and landing, the PrecisionHawk Lancaster was not granted an emergency certificate of authorization (COA), and the AirRobot flights were canceled due to concerns over privacy. Section 5 describes how the team returned and on April 23, 2014 flew an AirRobot AR100B quadrotor under an emergency COA with postprocessing by Precision Hawk producing a two-dimensional (2D) mosaic and 3D interactive reconstruction of 30– 40 acres of the moonscape with 48 min of flight time and 3 h of processing time on a laptop. The general performance of the SUAS, the lessons learned for operations, the human-robot ratio for the missions, and gaps and open research questions are discussed in Section 6. The article concludes in Section 7 that SUASs are costand time-effective for mudslide response. 2. Prior And Related Work SUAS use has been reported for 17 disasters, including the SR-530 mudslide incident and a subsequent 2014 mudslide in Collbran, CO. Deployments to 11 of the 17 disasters are analyzed in the book Disaster Robotics (Murphy, 2014): Hurricane Katrina, USA (2005); Hurricane Wilma, USA (2005); Berkman Plaza II Collapse, USA (2007); L’aqulia Earthquake, Italy (2009); Haiti Earthquake (2010); Christchurch Earthquake, New Zealand (2011); Tohoku Earthquake, Japan (2011); Fukushima Daiichi Nuclear Accident, Japan (2011); Evangelos Florakis Naval Base Explosion, Cyprus (2011); Thailand Floods (2011); and Finale Emilia Earthquake, Italy (2012). The deployment to Typhoon Morakot, Taiwan (2009) is reported in Adams & Friedland (2011). The deployments to Typhoon Haiyan, Philippines (2013) (University of Hawai’i, 2014; UH, 2014); the Boulder, CO floods, USA (2013) (9News, 2013); the Collbran, CO mudslide, USA (2014) (Yoanna, 2014); and the Serbia and Bosnia-Herzegovina floods (2014) (ICARUS, 2014) were reported in the media. The SR-530 flights differ from prior work. No SUAS deployments prior to the SR-530 event were for mudslides, which consist of a vertical scarp and horizontal deposits of Figure 1. SR-530 mudslide labeled with regions of interest. U s e o f S UA S f o r M u d s l i d e I n c i d e n t n e a r O s o , Wa s h i n g t o n 3 mud downslope. The 2011 Thailand Floods and the 2013 Typhoon Haiyan missions specifically looked at flooding. These deployments are especially relevant because flooding is a continuing consequence of the engagement of the Stillaguamish River in the SR-530 mudslide, and thus they are compared to the SR-530 deployment below. No pre-2014 deployment reported postprocessing for 3D reconstruction of terrain, although 2D mosaics were implied in University of Hawai’i (2014) and UH (2014). 2.1. Relevant CRASAR Deployments CRASAR deployed SUASs to five of the 17S UAS events (Katrina, Wilma, Berkman Plaza II, L’aquila, and Fukushima Daiichi) prior to the SR-530 mudslide. SUAS protocols developed by CRASAR were used by teams at two other events (Evangelos Florakis, Finale Emilia). CRASAR had also deployed unmanned ground robots to help search collaterally damaged houses at the La Conchita, CA Mudslide (2005). The La Conchita mudslide was significantly different from the SR350 slide in that it was a narrow slide of about 0.035 km2 with a claylike solid mud that supported the weight of responders and equipment. One of the recommendations from that deployment was to use robots to monitor the mudslide (Murphy & Stover, 2008); the SR-530 deployment is thus a logical extension of the 2005 deployment. 2.2. 2011 Thailand Flooding The 2011 Thailand Floods was the first event in which SUASs are known to have been specifically used for flood assessment. Siam UAV Industries deployed an eSUAV600 small electric-powered fixed wing with roughly a 2.5 m wing span and video cameras for 3 months in 2011 to assist the Thai government with the major flooding event (Srivaree-Ratana, 2012). The mission was to provide video of the water movement and the status of mitigation work to engineers and officials so that they could better control the flooding and evacuate the population. Siam UAV Industries obtained permission from the government to fly. The airspace was divided by the government into manned and unmanned regions in order to prevent possible collisions with news helicopters flying at low altitudes. The missions were successful, but lessons learned emphasized the need to annotate and curate the large amount of video, as geotagging was not sufficient. 2.3. 2013 Typhoon Haiyan A set of fixed-wing SUASs with video cameras was deployed by the University of Hawai’i Hilo to the 2013 Typhoon Haiyan (UH, 2014; University of Hawai’i, 2014) for general visual mapping surveys including the Aklan river system at Panay. The team used a custom-built fixed-wing SUAS with approximately a 1.5 m wing span. Work was directed by an organization called Skyeye associated with the Ateneo de Manila University, and the data were provided to local governments. The value of surveying the river system was to understand and prevent additional flooding. The UH Hilo team described that they added “first person view” [generally referred to as remote presence (Murphy & Burke, 2008; Tittle, Roesler, &Woods, 2002) in the cognitive engineering and human-robot interaction literature] so that the responders could actively manage missions and for general safety. The team also cited difficulties with the weather and with finding launch sites. Other SUAS platforms were reported in the media as being used at Typhoon Haiyan for general damage assessment and general mapping, but not flood assessment. Some notable examples are Team Rubicon, which deployed a Huginn X1 quadrotor with a camera and IR (Net Hope Center, 2014), and unspecified humanitarian relief organizations supporting Open Street Maps (OpenStreetMap, 2014). There was no discussion of flight altitudes, airspace regulations, or airspace deconfliction with manned assets operating in the same area, although one website reported that the Philippine Civil Aviation Authority had restricted flights to Tacloban City to relief effort flights only. 2.4. Similarities and Differences with SR-530 Deployments The SR-530 Mudslide deployments are most similar to the Typhoon Haiyan deployment. They shared a similar motivation to mitigate risk to citizens (and responders) from current and future flooding. Both the RWB and UH Hilo teams had difficulty in finding staging areas and were prevented by weather from flying some flights. In both cases, the platform and payload selection were made on the presumption of mapping missions, but the need for a remote presence emerged in the field. The SR-530 deployments are different from the Thailand Floods and Typhoon Haiyan deployments. Those deployments relied solely on fixed wing, not rotorcraft, and the visualizations appear to be limited to 2D mosaics versus 3D interactive reconstructions. The airspace in both cases was not restricted in the areas in which they appeared to be working, and thus it was open to news helicopters and other response workers, whereas the airspace in the SR-530 event was under a temporary flight restriction (TFR) barring entry of all aircraft except those authorized through the incident command process. The Asian deployments appeared to have deconflicted airspace either by explicitly sterilizing the airspace or through some informal method. 3. Premission Selection of Platforms On March 27, 2014, CRASAR received an invitation to fly SUASs in order to provide data for the geological and hydrological teams through the Snohomish County Sheriff Urban Search and Rescue Air Operations branch, facilitated by 4 M u r p h y e t a l . i n J o u r n a l o f F i e l d R o b o t i c s ( 2 0 1 6 ) the field innovation team (FIT), a disaster response nonprofit that delivers innovative solutions, real-time, to help first responders and disaster survivors. 3.1. Tactical and Strategic Mission Objectives The initial mission scope as described prior to arrival had three objectives, two of which required advanced data processing and visualization for strategic decision makers. The first objective was to aid tactical teams in anticipating and mitigating ongoing flooding by providing comprehensive imagery. The responders were concerned that the continuing rain in the region combined with the dynamically changing river course could lead to significant flooding. Flooding could impact other residents but also the responders working downslope on recovery, creating a second disaster. The second objective was to provide a rapid 3Dreconstruction of the site and then use that to create a 3D printed map of the terrain. FIT provided access to Autodesk as ReCap reality capture, a cloud-point image-based 3D modeling software, and an Objet 500 Connex 3D printer. The 3D printed map would complement a 3D graphics reconstruction and also serve as a physical artifact for the responders to work over together, thus offering a similar but different visualization capability. The third objective was to collect imagery over several days to anticipate further slide movement so as to protect the responders. Note that there was no mission to search for survivors or for victim recovery. 3.2. Platform Selection Following Disaster Robotics (Murphy, 2014), the choice of platforms from within the Roboticists Without Borders membership was based on four questions: What are the expected (and unarticulated) needs for the robot? What are the transportation and logistical arrangements? How will maintenance and repairs be conducted? and Are there any regulatory issues that must be considered? The answer to the fourth question was that all platforms would require an emergency COA from the FAA to fly. The first three questions produced the following set of criteria: • Prior use for geospatial missions. Reliable platforms that had existing payloads and demonstrated post-processing for accurate, georeferenced 2D tiling and 3D reconstruction were needed. • Rapid coverage and reconstruction. The SR-350 slide was approximately 2.5 km2. The general expectation was that the entire process—flight and postprocessing— should occur within 1 day, preferably within one 12-h shift so that the data could be used for planning the next day’s activities. • Portability. It was expected that SUASs and operator control stations would have to be manually carried to a suitable launch site inside the disaster zone. • Ability to fly in regional weather. Washington State is subject to rain. While SUASs generally are not permitted to operate in rain due to the lack of visibility, it is desirable to be able to operate in light rain so that the platform could get wet as it was returning home. CRASAR invited Insitu and PrecisionHawk to join the deployment under the Roboticists Without Borders program, where they donate their time and travel costs, and they absorb any damage or loss to the platform. The expectation was to bring two complementary fixed-wings, namely the Insitu Scan Eagle and the PrecisionHawk Lancaster, with geospatial sensing and postprocessing capabilities, and to bring the CRASAR AirRobot AR100B quadrotors as a backup (see Figure 2). It should be noted that over two dozen SUAS systems are commercially available that support geospatial missions; these two can be considered representative of the emerging industry. The AirRobot AR100B is a man-portable rotorcraft weighing 1.8 kg with vertical takeoff and landing with operations in up to 15 knots, and flight durations of 8–20 min within about 3 km. It is typically used at altitudes between 9 and 122 m AGL. It is used primarily for military or border security applications, where the real-time low-resolution imagery from a RGB camera is used to investigate situations or track activity. The Panasonic Lumix 10 megapixel camera transmits a 640 × 480 viewfinder image over 802.11 b/g/n in real time; this low-resolution viewfinder image is used for teleoperation. The AR100B can take manually high-resolution still imagery, but software upgrades for automated image collection for use with postprocessing software was not available at the time of the deployment. Other payloads, such as fused video and thermal imaging, were not available on loan from the manufacturer. The platform can operate in a light rain. The AirRobot was chosen as a backup platform because it could be launched vertically and because it could provide responders with tactical, on-demand oversight of the general area. The Insitu Scan Eagle is a fixed-wing UAS with a wing span of 3.1 m and a weight of 14 kg. It requires a short runway to launch and land and is supported by three tractortrailer units. It was chosen despite its larger size, weight, and staging needs, because it is used extensively by the U.S. Geological Survey, it is arguably the best known UAS for geospatial application, and because the company is based in nearby Oregon and could respond quickly. The PrecisionHawk Lancaster is a fixed-wing man-portable SUAS with a 1.2 m wing span, weighing 2.5 kg. It is handlaunched with operations in up to 25 knots of wind and the platform belly-lands as opposed to having landing gear unless landing in water, at which time floats can be employed. It can be landed either automatically or manually. The Lancaster is primarily used for agricultural and terrain mapping at altitudes of 30.5–183 m above ground level using video, LIDAR, or thermal payloads. It has flight durations of approxU s e o f S UA S f o r M u d s l i d e I n c i d e n t n e a r O s o , Wa s h i n g t o n 5 imately 60 min and can map a 2.6 km2 area in under 2 h. The reconstructions can be generated from either video or linescanning LIDAR payloads, with 3 cm per pixel processed in 3–72 h. The LIDAR payload was not available for the mudslide deployments. The Lancaster was chosen because of its high degree of portability and flexibility in staging, postprocessing software for terrain reconstruction, and the ability to fly in light rain. 4. March 2014 Deployment The team assembled on March 28, 2014, and demobilized on March 30. The team was directed by Robin Murphy (CRASAR), with Brittany Duncan (CRASAR) as the pilotin-command. Tyler Collins was the lead pilot for PrecisionHawk, with Pat Lohman as field support. Kevin Cole and Travis Cieloha were the lead pilots for Insitu. Frank Sanborn (FIT) served as the liaison with the incident command management team but had no direct responsibilities for the SUAS. Friday, March 28 was spent waiting for the three teams to arrive and scouting staging areas that could serve as takeoff and landing zones and provide visibility for lineof-sight operations. The deployment resulted in zero flights due to environmental constraints eliminating the Insitu and county concerns over safety and privacy, despite meeting FAA regulations. 4.1. Environmental Constraints on Operations The Insitu Scan Eagle demobilized on the afternoon of March 28 due to a lack of a suitable staging area within the TFR. The temporary heliport at Skagland that was being used by manned helicopters had sufficient space and access, but there was a possibility of radio interference between the Black Hawks and Scan Eagle, as well as complicated coordination issues. No other site was found. The CRASAR and PrecisionHawk representatives scouted for a location that could be used by both platforms. A temporary emergency access south of SR-530 was ruled out due to radio-frequency interference from overhead power lines [Figure 3(a)]. It also would have required over a 400 m hike over a steep muddy hill; see Figure 3(b). A meadow at the Shunn property off Whitman Road was identified as the best option. It would not require a hike but it did require permission from the landowner, a four-wheel-drive vehicle to go off road, and there was a danger of a secondary slide. The location would also necessitate two safety officers in order to maintain constant line of sight with the SUAS. One would be with the flight team on the west side of the slope, and the second would be stationed on the south side of the slide and communicate with the team via radio. 4.2. Mission Objectives The mission objectives in order of priority are given below, with the areas for each mission shown in Figure 4. The refinement of missions introduced one surprise: that the quadrotor had a mission and was no longer strictly a backup aircraft. The expectation was to use the PrecisionHawk Lancaster flying at 137 m AGL for priorities 2–4, with follow-up flights if needed by the AR100B rotorcraft to investigate areas of interest from 30.5 m or less. However, for priority 1, the “hover and stare” capability of the AR100B was considered essential in allowing the Army Corps of Engineers to access the flow patterns and general movement of the water. A flight might be able to meet multiple objectives. • Priority 1: Riverbed assessment (blue). Washington Task Force 1, the state team conducting rescue and recovery operations, wanted the river bed cleared so the pond area would drain and they could search the waterlogged area. Low-altitude, high-resolution data would aid hydrologists in making decisions on where blockages are and how to clear them. • Priority 2: High resolution of lower slide (yellow). The geologists wanted a better understanding of the scarp, particularly at the toe. Low-altitude, high-resolution data would aid in identifying potential problem areas that would lead to more slides or make further changes in the river. • Priority 3: High resolution of cliff face/upper slide (green). Washington Task Force 1 wanted to project secondary slides because even a small slide could impact search (a) (b) (c) Figure 2. Roboticists Without Borders platforms selected for deployment: (a) the AirRobot AR100B, (b) the Insitu Scan Eagle, and (c) the Preci-