Tag Archives: out

#439882 Robot umpires are coming to baseball. ...

Baseball fans know the bitter heartbreak of calls that don't go their way—especially, a ball that should've been a strike. And, with advances in technology including computer vision, artificial intelligence, and the ubiquity of Wi-Fi, it would be easier than ever for baseball officials to replace humans with robotic umpires. Continue reading

Posted in Human Robots

#439543 How Robots Helped Out After the Surfside ...

Editor's Note: Along with Robin Murphy, the authors of this article include David Merrick, Justin Adams, Jarrett Broder, Austin Bush, Laura Hart, and Rayne Hawkins. This team is with the Florida State University's Disaster Incident Response Team, which was in Surfside for 24 days at the request of Florida US&R Task 1 (Miami Dade Fire Rescue Department).

On June 24, 2021, at 1:25AM portions of the 12 story Champlain Towers South condominium in Surfside, Florida collapsed, killing 98 people and injuring 11, making it the third largest fatal collapse in US history. The life-saving and mitigation Response Phase, the phase where responders from local, state, and federal agencies searched for survivors, spanned June 24 to July 7, 2021. This article summarizes what is known about the use of robots at Champlain Towers South, and offers insights into challenges for unmanned systems.

Small unmanned aerial systems (drones) were used immediately upon arrival by the Miami Dade Fire Rescue (MDFR) Department to survey the roughly 2.68 acre affected area. Drones, such as the DJI Mavic Enterprise Dual with a spotlight payload and thermal imaging, flew in the dark to determine the scope of the collapse and search for survivors. Regional and state emergency management drone teams were requested later that day to supplement the effort of flying day and night for tactical life-saving operations and to add flights for strategic operations to support managing the overall response.

View of a Phantom 4 Pro in use for mapping the collapse on July 2, 2021. Two other drones were also in the airspace conducting other missions but not visible. Photo: Robin R. Murphy
The teams brought at least 9 models of rotorcraft drones, including the DJI Mavic 2 Enterprise Dual, Mavic 2 Enterprise Advanced, DJI Mavic 2 Zoom, DJI Mavic Mini, DJI Phantom 4 Pro, DJI Matrice 210, Autel Dragonfish, and Autel EVO II Pro plus a tethered Fotokite drone. The picture above shows a DJI Phantom 4 Pro in use, with one of the multiple cranes in use on the site visible. The number of flights for tactical operations were not recorded, but drones were flown for 304 missions for strategic operations alone, making the Surfside collapse the largest and longest use of drones recorded for a disaster, exceeding the records set by Hurricane Harvey (112) and Hurricane Florence (260).

Unmanned ground bomb squad robots were reportedly used on at least two occasions in the standing portion of the structure during the response, once to investigate and document the garage and once on July 9 to hold a repeater for a drone flying in the standing portion of the garage. Note that details about the ground robots are not yet available and there may have been more missions, though not on the order of magnitude of the drone use. Bomb squad robots tend to be too large for use in areas other than the standing portions of the collapse.

We concentrate on the use of the drones for tactical and strategic operations, as the authors were directly involved in those operations. It offers a preliminary analysis of the lessons learned. The full details of the response will not be available for many months due to the nature of an active investigation into the causes of the collapse and due to privacy of the victims and their families.
Drone Use for Tactical Operations
Tactical operations were carried out primarily by MDFR with other drone teams supporting when necessary to meet the workload. Drones were first used by the MDFR drone team, which arrived within minutes of the collapse as part of the escalating calls. The drone effort started with night operations for direct life-saving and mitigation activities. Small DJI Mavic 2 Enterprise Dual drones with thermal camera and spotlight payloads were used for general situation awareness to help responders understand the extent of the collapse beyond what could be seen from the street side. The built-in thermal imager was used but did not have the resolution and was unable to show details as much of the material was the same temperature and heat emissions were fuzzy. The spotlight with the standard visible light camera was more effective, though the view was constricted. The drones were also used to look for survivors or trapped victims, help determine safety hazards to responders, and provide task force leaders with overwatch of the responders. During daylight, DJI Mavic Zoom drones were added because of their higher camera resolution zoom. When fires started in the rubble, drones with a streaming connection to bucket truck operators were used to help optimize position of water. Drones were also used to locate civilians entering the restricted area or flying drones to taking pictures.

In a novel use of drones for physical interaction, MDFR squads flew drones to attempt to find and pick up items in the standing portion of the structure with immense value to survivors.

As the response evolved, the use of drones was expanded to missions where the drones would fly in close proximity to structures and objects, fly indoors, and physically interact with the environment. For example, drones were used to read license plates to help identify residents, search for pets, and document belongings inside parts of the standing structure for families. In a novel use of drones for physical interaction, MDFR squads flew drones to attempt to find and pick up items in the standing portion of the structure with immense value to survivors. Before the demolition of the standing portion of the tower, MDFR used a drone to remove an American flag that had been placed on the structure during the initial search.

Drone Use for Strategic Operations

An orthomosiac of the collapse constructed from imagery collected by a drone on July 1, 2021.
Strategic operations were carried out by the Disaster Incident Research Team (DIRT) from the Florida State University Center for Disaster Risk Policy. The DIRT team is a state of Florida asset and was requested by Florida Task Force 1 when it was activated to assist later on June 24. FSU supported tactical operations but was solely responsible for collecting and processing imagery for use in managing the response. This data was primarily orthomosiac maps (a single high resolution image of the collapse created from stitching together individual high resolution imagers, as in the image above) and digital elevation maps (created from structure from motion, below).

Digital elevation map constructed from imagery collected by a drone on 27 June, 2021.Photo: Robin R. Murphy
These maps were collected every two to four hours during daylight, with FSU flying an average of 15.75 missions per day for the first two weeks of the response. The latest orthomosaic maps were downloaded at the start of a shift by the tactical responders for use as base maps on their mobile devices. In addition, a 3D reconstruction of the state of the collapse on July 4 was flown the afternoon before the standing portion was demolished, shown below.

GeoCam 3D reconstruction of the collapse on July 4, 2021. Photo: Robin R. Murphy
The mapping functions are notable because they require specialized software for data collection and post-processing, plus the speed of post-processing software relied on wireless connectivity. In order to stitch and fuse images without gaps or major misalignments, dedicated software packages are used to generate flight paths and autonomously fly and trigger image capture with sufficient coverage of the collapse and overlap between images.

Coordination of Drones on Site
The aerial assets were loosely coordinated through social media. All drones teams and Federal Aviation Administration (FAA) officials shared a WhatsApp group chat managed by MDFR. WhatsApp offered ease of use, compatibility with everyone's smartphones and mobile devices, and ease of adding pilots. Ease of adding pilots was important because many were not from MDFR and thus would not be in any personnel-oriented coordination system. The pilots did not have physical meetings or briefings as a whole, though the tactical and strategic operations teams did share a common space (nicknamed “Drone Zone”) while the National Institute of Standards and Technology teams worked from a separate staging location. If a pilot was approved by MDFR drone captain who served as the “air boss,” they were invited to the WhatsApp group chat and could then begin flying immediately without physically meeting the other pilots.

The teams flew concurrently and independently without rigid, pre-specified altitude or area restrictions. One team would post that they were taking off to fly at what area of the collapse and at what altitude and then post when they landed. The easiest solution was for the pilots to be aware of each others' drones and adjust their missions, pause, or temporarily defer flights. If a pilot forgot to post, someone would send a teasing chat eliciting a rapid apology.
Incursions by civilian manned and unmanned aircraft in the restricted airspace did occur. If FAA observers or other pilots saw a drone flying that was not accounted for in the chat, i.e., that five drones were visible over the area but only four were posted, or if a drone pilot saw a drone in an unexpected area, they would post a query asking if someone had forgotten to post or update a flight. If the drone remained unaccounted for, the FAA would assume that a civilian drone had violated the temporary flight restrictions and search the surrounding area for the offending pilot.
Preliminary Lessons LearnedWhile the drone data and performance is still being analyzed, some lessons learned have emerged that may be of value to the robotics, AI, and engineering communities.
Tactical and strategic operations during the response phase favored small, inexpensive, easy to carry platforms with cameras supporting coarse structure from motion rather than larger, more expensive lidar systems. The added accuracy of lidar systems was not needed for those missions, though the greater accuracy and resolution of such systems were valuable for the forensic structural analysis. For tactical and strategic operations, the benefits of lidar was not worth the capital costs and logistical burden. Indeed, general purpose consumer/prosumer drones that could fly day or night, indoors and outdoors, and for both mapping and first person view missions were highly preferred over specialized drones. The reliability of a drone was another major factor in choosing a specific model to field, again favoring consumer/prosumer drones as they typically have hundreds of thousand hours of flight time more than specialized or novel drones. Tethered drones offer some advantages for overwatch but many tactical operations missions require a great deal of mobility. Strategic mapping necessitates flying directly over the entire area being mapped.

While small, inexpensive general purpose drones offered many advantages, they could be further improved for flying at night and indoors. A wider area of lighting would be helpful. A 360 degree (spherical) area of coverage for obstacle avoidance for working indoors or at low altitudes and close proximity to irregular work envelopes and near people, especially as night, would also be useful. Systems such as the Flyability ELIOS 2 are designed to fly in narrow and highly cluttered indoor areas, but no models were available for the immediate response. Drone camera systems need to be able to look straight up to inspect the underside of structures or ceilings. Mechanisms for determining the accurate GPS location of a pixel in an image, not just the GPS location of the drone, is becoming increasing desirable.
Other technologies could be of benefit to the enterprise but face challenges. Computer vision/machine learning (CV/ML) for searching for victims in rubble is often mentioned as a possible goal, but a search for victims who are not on the surface of the collapse is not visually directed. The portions of victims that are not covered by rubble are usually camouflaged with gray dust, so searches tend to favor canines using scent. Another challenge for CV/ML methods is the lack of access to training data. Privacy and ethical concerns poses barriers to the research community gaining access to imagery with victims in the rubble, but simulations may not have sufficient fidelity.
The collapse supplies motivation for how informatics research and human-computer interaction and human-robot interaction can contribute to the effective use of robots during a disaster, and illustrates that a response does not follow a strictly centralized, hierarchical command structure and the agencies and members of the response are not known in advance. Proposed systems must be flexible, robust, and easy to use. Furthermore, it is not clear that responders will accept a totally new software app versus making do with a general purpose app such as WhatsApp that the majority routinely use for other purposes.
The biggest lesson learned is that robots are helpful and warrant more investment, particular as many US states are proposing to terminate purchase of the very models of drones that were so effective over cybersecurity concerns.
However, the biggest lesson learned is that robots are helpful and warrant more investment, particular as many US states are proposing to terminate purchase of the very models of drones that were so effective over cybersecurity concerns. There remains much to work to be done by researchers, manufacturers, and emergency management to make these critical technologies more useful for extreme environments. Our current work is focusing on creating open source datasets and documentation and conducting a more thorough analysis to accelerate the process.

Value of Drones The pervasive use of the drones indicates their implicit value to responding to, and documenting, the disaster. It is difficult to quantify the impact of drones, similar to the difficulties in quantifying the impact of a fire truck on firefighting or the use of mobile devices in general. Simply put, drones would not have been used beyond a few flights if they were not valuable.
The impact of the drones on tactical operations was immediate, as upon arrival MDFR flew drones to assess the extent of the collapse. Lighting on fire trucks primarily illuminated the street side of the standing portion of the building, while the drones, unrestricted by streets or debris, quickly expanded situation awareness of the disaster. The drones were used optimize placement of water to suppress the fires in the debris. The impact of the use of drones for other tactical activities is harder to quantify, but the frequent flights and pilots remaining on stand-by 24/7 indicate their value.
The impact of the drones on strategic operations was also considerable. The data collected by the drones and then processed into 2D maps and 3D models became a critical part of the US&R operations as well as one part of the nascent investigation into why the building failed. During initial operations, DIRT provided 2D maps to the US&R teams four times per day. These maps became the base layers for the mobile apps used on the pile to mark the locations of human remains, structural members of the building, personal effects, or other identifiable information. Updated orthophotos were critical to the accuracy of these reports. These apps running on mobile devices suffered from GPS accuracy issues, often with errors as high as ten meters. By having base imagery that was only hours old, mobile app users where able to 'drag the pin' on the mobile app to a more accurate report location on the pile – all by visualizing where they were standing compared to fresh UAS imagery. Without this capability, none of the GPS field data would be of use to US&R or investigators looking at why the structural collapse occurred. In addition to serving a base layer on mobile applications, the updated map imagery was used in all tactical, operational, and strategic dashboards by the individual US&R teams as well as the FEMA US&R Incident Support Team (IST) on site to assist in the management of the incident.
Aside from the 2D maps and orthophotos, 3D models were created from the drone data and used by structural experts to plan operations, including identifying areas with high probabilities of finding survivors or victims. Three-dimensional data created through post-processing also supported the demand for up-to-date volumetric estimates – how much material was being removed from the pile, and how much remained. These metrics provided clear indications of progress throughout the operations.
Acknowledgments Portions of this work were supported by NSF grants IIS-1945105 and CMMI- 2140451. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
The authors express their sincere condolences to the families of the victims. Continue reading

Posted in Human Robots

#439443 This Robot Taught Itself to Run, Then ...

In the last few months, robots have learned some pretty cool new skills, including performing a sweet coordinated dance routine and making pizzas from start to finish. Now there’s another accomplishment to add to the list: a bipedal robot named Cassie just ran a 5K.

Made by Agility Robotics, which was spun out of Oregon State University, Cassie was developed using a $1 million grant from DARPA. The robot is basically a pair of mechanical legs with a battery pack sitting on top. Thanks to the design of its hip joints, its legs can move forward, backward, or side to side.

Earlier this year, a group of students at Berkeley used machine learning to teach Cassie to walk. But making the leap from walking to running wasn’t as straightforward as you might think. To us, running is just a faster version of walking, and we don’t often consider the various skills and brain regions that go into even a short jog around the neighborhood.

Our core muscles engage to help keep us balanced as we’re in constant motion. Our vision scans the area in front of us for obstacles to avoid, changing course as necessary. Our heart rate kicks up a few notches, and our respiratory system regulates our breathing.

Granted, it’s a little different for a robot, since they don’t have lungs or a heart. But they do have a “brain” (software), “muscles” (hardware), and “fuel” (a battery), and these all had to work together for Cassie to be able to run.

The brunt of the work fell to the brain—in this case, a machine learning algorithm developed by students at Oregon State University’s Dynamic Robotics Laboratory. Specifically, they used deep reinforcement learning, a method that mimics the way humans learn from experience by using a trial-and-error process guided by feedback and rewards. Over many repetitions, the algorithm uses this process to learn how to accomplish a set task. In this case, since it was trying to learn to run, it may have tried moving the robot’s legs varying distances or at distinct angles while keeping it upright.

Once Cassie got a good gait down, completing the 5K was as much a matter of battery life as running prowess. The robot covered the whole distance (a course circling around the university campus) on a single battery charge in just over 53 minutes, but that did include six and a half minutes of troubleshooting; the computer had to be reset after it overheated, as well as after Cassie fell during a high-speed turn. But hey, an overheated computer getting reset isn’t so different from a human runner pausing to douse their head and face with a cup of water to cool off, or chug some water to rehydrate.

Cassie isn’t the first two-legged robot to run. Honda’s Asimo robot had a slow jog down in 2004, and Boston Dynamics’ Atlas bot looks (sort of frighteningly) like a person when it runs, moving its arms in coordination with its legs. But it is notable that Cassie taught itself to run, as it shows off machine learning’s potential in robotic systems.

And this feat is just the beginning. “The students combined expertise from biomechanics and existing robot control approaches with new machine learning tools,” said Jonathan Hurst, a robotics professor who co-founded Agility in 2017. “This type of holistic approach will enable animal-like levels of performance. It’s incredibly exciting.”

Image Credit: Agility Robotics/Oregon State University Dynamic Robotics Laboratory Continue reading

Posted in Human Robots

#439439 Swarms of tiny dumb robots found to ...

A team of researchers affiliated with several institutions in Europe has found that swarms of tiny dumb vibrating robots are capable of carrying out sophisticated actions such as transporting objects or squeezing through tunnels. In their paper published in the journal Science Robotics, the group describes experiments they conducted with tiny dumb robots they called “bugs.” Continue reading

Posted in Human Robots

#439125 Baubot comes out with two new robots to ...

Despite artificial intelligence and robotics adapting to many other areas of life and the work force, construction has long remained dominated by humans in neon caps and vests. Now, the robotics company Baubot has developed a Printstones robot, which they hope to supplement human construction workers onsite. Continue reading

Posted in Human Robots