DV1.1

A collaborative scanning mission between an RMAX and a LinkQuad. A scan area is divided into two non-overlapping sub-areas for the two platforms.

DV1.2

Part 1 of a collaborative scanning mission between an RMAX and a LinkQuad. The RMAX builds an overview occupancy map which is used by the LinkQuad to plan paths and film a building structure from all sides.

DV1.3

Part 2 of the mission. Filming of the façades by the LinkQuad.

DV1.4

Raw sensor data collected by the RMAX using a laser range finder, a color camera, and a thermal camera.

DV2.1

A collaborative and concurrent scanning mission between a DJI M600 and a DJI M100. Two non-overlapping regions are scanned at the same time.

DV2.2

A DJI M100 performs a laser scan of a part of a region previously scanned by the DJI M600.

DV2.3

An example mission where three DJI M100 platforms first pick up supplies from three locations and deliver them to different locations. The actual attaching/delivery mechanism was not ready at this time.

DV2.4

Simulation of a mission described above.

DV2.5

Data generated by a M600 during a laser scanning mission.

DV2.6

A mission involving two DJI M100 platforms and a SAAB Kockums Piraya autonomous boat. One M100 platform observes the mission area while the second UAV is leashed to the boat i.e. it autonomously follows it. The boat navigates to a sequence of waypoints. The SAFE user interface is used to manage the mission.

DV3.1

A demonstration of the vision detector/tracker integrated with the DJI M100 platform. The UAV is stationary, the gimbal follows the tracked target.

DV3.2

A demonstration of the integrated detection+tracking algorithm in comparison to the pure tracking algorithm. Long term occlusions are successfully dealt with by the integrated approach, in contrast to the pure tracking approach.

DV3.3

Similar to DV3.2, but now addressing out-of-view frames. Also, cases where the target leaves the field of view require the integrated approach.

DV4.1

A mission involving the Husky robot delivering a package to the RMAX. A simulated failure makes the ground robot ask a human for help.

DV5.1

An example mission performed during development of the CommKit delivery system.

DV5.2

An example mission performed during development of the CommKit delivery system.

DV5.3

An example mission performed during development of the CommKit delivery system. Onboard camera view.

DV5.4

A full mission demonstrating the CommKit delivery.

DV6.1

An example mission demonstrating the usage of a human body classifier in a search and rescue mission.

DV7.1

A mission demonstrating further development of the virtual leashing functionality shown in DM1.4 / DM3.9. The vision detector/tracker is integrated with the DJI M100 platform. The UAV follows the tracked person, while altitude is kept constant.

DV8.1

Deep learning quadcopter control with risk-aware active learning.

DV8.2

Model Predictive Control for Stochastic Collision Avoidance using Bayesian Policy Optimisation.

DV9.1,
DV9.2,
DV9.3

Examples of flight functionalities implemented using the method described in section Control interfaces for mixed initiative interaction in the midterm report. The method allows for implementing a wide range of flight capabilities (takeoff, landing, waypoint following, velocity control etc.) as well as easy switching between them to enable a high level of interactivity.

DV9.4

Learning Intensity Maps for Informed Robotic Search

CAVOID

This video demonstrates real-time decision making under uncertainty applied to safe navigation of a UAV platform. A LiDAR sensor is used to detect in real-time the locations of static and dynamic obstacles in order to navigate (with safety guarantees) in the environment where non-cooperating third parties are present.

AID1

This video demonstrates a mission where a UAV platform autonomously delivers two medical kits to locations of previously identified human victims. The two medkits are delivered in sequence.

BATT1

This video demonstrates coping with contingencies during an autonomous continuous observation mission. Three simulated platforms choose and observe geographical locations. In case a low-battery signal is detected, the UAV in question requests help from a human to replace a battery at a supply depot. After the battery replacement is done, the UAV rejoins the mission.

INAIR1

This video demonstrates functionalities used for a mission requiring delivery of a small UAV platform by another one. When a request is made for a small UAV by a rescuer, a human is asked by a carrier UAV to attach the small platform. After that, the platform is delivered and deployed in the vicinity of the rescuer to be leashed to him/her. The mission is performed in simulation.

MIT1

This video demonstrates rich interaction capabilities between desktop/VR interfaces and autonomous UAV systems. This version demonstrates interaction with a simulated UAV platform.

MIT2

This video demonstrates rich interaction capabilities between desktop/VR interfaces and autonomous UAV systems. This version demonstrates interaction with a real UAV platform (extension of work presented in the video above).

LAND

This video demonstrates a vision-based functionality for a UAV platform to identify, track, and land on an object whose image is provided to the system at runtime.

AID2

This video demonstrates an algorithm for detecting and geolocating objects of diverse types for the purpose of search and rescue missions. Video signals are distributively processed using available cloud or local computational resources. The detected objects in the video streams are geolocated and fused into a map of salient locations.

AID3

This video describes in more detail the functionality for object detection and geolocation for search and rescue missions (extended version of video above).

BATT2

This video demonstrates autonomous continuous observation mission performed by a team of three UAVs. Functionalities used in the mission include a concept of Team Task Queues and autonomous joining and leaving the mission while it is being executed. Additionally, the battery replacement functionalities are shown. Note: similar mission scenario has been demonstrated during WARA-PS 2021 Annual Workshop.

INAIR2

This video demonstrates a proof-of-concept functionality of in-flight releasing of a small-scale UAV platform from another one. Here a Ryze Tello micro-UAV platform (80g of weight) is released in-air from a larger DJI Matrice M100 platform (2.5kg of weight).

GEST

This video demonstrates commanding of a UAV platform by a rescuer using vision-based gesture detection. Several predefined gestures are used by a human operator to: get a platform's attention, takeoff, initiate vision-based leashing, adjust relative position between the UAV and the rescuer, and land. The gesture recognition system works in real-time and is coupled to a simulated UAV platform for the purpose of this experiment.

SCQL

This video demonstrates the use of SymbiCloud Query Language. Two queries for Lidar data are executed for different areas. The queries result in the automatic generation and execution of exploration missions. In the second query, the selected region overlaps with the region where data was already acquired; thus, a mission is generated that only considers acquiring the missing data. This shows the ability to reuse existing information in later queries.