DS0705 - Fondements du numérique

Sensor-Based Flying Multi-Robot System – SenseFly

SenseFly

Advancing the state-of-the-art in sensor-based multi-robot applications, with a special focus on quadrotor UAVs and onboard camera. The main challenges are in implementing decentralized cooperative estimation and formation control schemes by only using onboard sensing (vision) and communicaiton, and able to cope with the onboard sensing limitations

Decentralized formation control and localization for a quadrobot group with onboard cameras

The main challenges are in implementing decentralized cooperative estimation and formation control schemes by only using onboard sensing (vision) and communicaiton, and able to cope with the onboard sensing limitations. We chose quadrotor UAVs because of their low cost, agility, and pervasiveness in 3D space.<br />As a concrete application of the project results, SenseFly will aim at exploiting the UAV group as a versatile and mobile localization system able to provide localization services to other robots in visibility of some of the UAVs by only exploiting the group local sensing and communication, and despite the limitations of the individual onboard sensors (e.g., pair-wise occlusions or limited visibility among team members). This will effectively result in an autonomous flying mobile localization system ready to be deployed in complex/cluttered indoor/outdoor non-engineered environments, thus freeing from the need of relying on complex external (Vicon-like) facilities or possibly unreliable services (GPS) by solely relying on the group skills.

The goal of the SenseFly project is to advance the state-of-the-art in sensor-based cooperative estimation and control of multi-robot systems with a particular focus on the quadrotor UAV platform, which is a widespread mobile robotic platform with low cost, high agility and pervasiveness in 3D space. Multi-robot systems (and, in particular multi-UAV systems) are of high interest for many tasks spanning autonomous search and rescue, firefighting, exploration and navigation in dangerous/remote areas. The main idea is that proper coordination of many simple robots can lead to the fulfillment of arbitrarily complex tasks in a robust and flexible way. However, to date, most implementations of multi-robot systems heavily rely on centralized facilities/aids, such as GPS (in outdoor conditions), mapping/SLAM in GPS-denied environments, or the extensive use of motion tracking systems (e.g., Vicon) for validation in lab conditions. In this respect, the goal of SenseFly is to enable a team of multiple quadrotor UAVs able to accomplish a mission by only resorting to its “own skills”, that is, local sensing (mainly IMU and vision) and local communication with neighboring robots. The main challenges to be addressed in the project are the use of onboard (local) sensing and communication for allowing a cooperative localizaiton and safe navigation of a group of quadrotor UAVs without external aids

We have implemented and tested several cooperative strategies for multiple quadrotor UAVs able to self-localize and navigate from onboard bearing measurements (what can be retrieved from onboard cameras). These theoretical advancements have also been validated in real experiments with five quadrotor UAVs

We are now addressing the issue of segmenting (in robust way) the other UAVs during flight, with the associated identity (needed by the control/estimation algorithms). We are also considering the issue of detecting obstacles in the scene

1. D. Zelazo, P. Robuffo Giordano, A. Franchi. Bearing-Only Formation Control Using an SE(2) Rigidity Theory. In IEEE Conf. on Decision and Control, CDC 2015, Osaka, Japan, December 2015
2. F. Schiano, A. Franchi, D. Zelazo, P. Robuffo Giordano. A Rigidity-Based Decentralized Bearing Formation Controller for Groups of Quadrotor UAVs. In IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, IROS'16, Pages 5099-5106, Daejeon, Korea, October 2016
3. R. Spica, P. Robuffo Giordano. Active Decentralized Scale Estimation for Bearing-Based Localization. In IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, IROS'16, Pages 5084-5091, Daejeon, Korea, October 2016
4. F. Schiano, P. Robuffo Giordano. Bearing Rigidity Maintenance for Formations of Quadrotor UAVs. In IEEE Int. Conf. on Robotics and Automation, ICRA'17, Singapore, May 2017

The last decade has witnessed a growing interest in multi-robot applications, based on the idea that proper coordination of many simple robots can lead to the fulfilment of arbitrarily complex tasks in a robust (to single robot failures) and highly flexible way. Autonomous search and rescue, firefighting, exploration and intervention in dangerous or inaccessible areas are some of the most promising multi-robot applications.
A fundamental requirement when devising motion control strategies for multiple robots is to rely on only relative measurements w.r.t. other robots or the environment, as for example relative distances, bearings or positions. In fact, these can be obtained from direct onboard sensing, and are thus free from the presence of global localization modules such as GPS or SLAM algorithms, or any other form of centralized localization system. Similarly, when exploiting a communication medium for exchanging information among robots, decentralized solutions requiring only local and 1-hop information are always preferred because of their higher tolerance to faults and inherent lower computational and communication loads.
However, in practice real implementations of multi-robot applications are seldom decentralized and/or based on sole onboard sensing: typically, while the design stage is made decentralized and dependent on only relative information, the actual realizations are still strongly relying, implicitly or explicitly, on centralized facilities and absolute measurements. This is, for example, the case of all those implementations exploiting an external visual tracking system, such as Vicon (or even GPS), which, by tracking all the robots in the scene, allows mimicking the presence of onboard sensing out of a global/centralized overview of the group. Obviously, these solutions are appropriate for ease of development, debugging and testing, or for applications designed for well-defined spaces/environments (e.g., indoor factory floors or external outdoor environments covered by GPS). However, they obviously lack the flexibility for being deployed in real-world unstructured environments (e.g., indoor collapsed buildings, in highly crowded spaces, underwater or underground).

The goal of this project is therefore to substantially advance the state-of-the-art in the direction of overcoming the above-mentioned issues by employing a team of Micro Aerial Vehicles (MAVs) as robotic platform, with a particular focus on quadrotor MAVs (a widespread and easily customizable mobile robotic platform with low cost, high agility and pervasiveness in 3D space). The project will address the design and implementation of fully decentralized and sensor-based group behaviors by only resorting to onboard sensing (mainly cameras and IMU) and local communication (e.g., bluetooth communication, wireless networks). Topics such as individual flight control, formation control robust against sensor limitations (e.g., limited field of view, occlusions), distributed estimation of relative positions/bearings from local sensing, maintenance of architectural properties of a multi-MAV formation will be touched by the project.
As final demonstrator of the developed technology, the project will exploit the MAV group as a versatile and mobile localization system able to provide localization services to other robots in visibility of some of the MAVs (e.g., some ground robots). The MAVs will not be bound to keep a given specified formation, but they will freely arrange themselves in space according to any internal criterium, e.g., for enhancing the localization accuracy of the tracked ground robots, or for coping with their unavoidable sensory limitations (limited visibility, occlusions). This will effectively result in an autonomous flying mobile localization system ready to be deployed in indoor (and possibly outdoor) environments, freeing from the need of relying on complex external (Vicon-like) facilities or possibly unreliable services (GPS).

Project coordination

Paolo Robuffo Giordano (Centre National de la Recherche Scientifique / Institut de Recherche en Informatique et systèmes Aléatoires)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

CNRS/IRISA Centre National de la Recherche Scientifique / Institut de Recherche en Informatique et systèmes Aléatoires

Help of the ANR 362,960 euros
Beginning and duration of the scientific project: September 2014 - 36 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter