The Drone revolution is running faster and faster. However, making a fully autonomous drone able to hover accurately near an object of interest, to explore its environment with or without GPS, to navigate in a cluttered environment without map and to avoid obstacles in a dynamic environment is still a great challenge for drone manufacturers and scientists working on aerial robots. This workshop provides participants with the unique opportunity to gain an overview of advanced developments for drones and aerial robots in terms of new sensors, autopilot and navigation strategies.
Participants can directly interact with the lead developers, dive into detailed discussions and meet other researchers interested in potential collaboration.
2:00 to 2:35 PM
1 Keynote : Davide Scaramuzza (Univ. of Zurich) (30 min + 5 min)
2:35 to 3:35 PM – Industrial developments for drones
- Sensefly : Jean-Christophe Zufferey (15 min + 5 min)
- Novadem : Sébastien Courroux (15 min + 5 min)
- L’avion jaune : Tristan Allouis (15 min + 5 min)
Coffee break – 10 minutes
3:45 to 5 PM : Academic development for aerial robots
- Antonio Franchi (LAAS-CNRS) (20 min + 5 min)
- Augustin Manecy (Gipsa-lab & ISM, now at ONERA Toulouse) (20 min + 5 min)
- Martial Sanfourche (ONERA Palaiseau) (20 min + 5 min)
Davide Scaramuzza (Univ. of Zurich)
Towards Robust and Safe Autonomous Drones
With drones becoming more and more popular, safety is a big concern. A critical situation occurs when a drone temporarily loses its GPS position information, which might lead it to crash. This can happen, for instance, when flying close to buildings where GPS signal is lost. In such situations, it is desirable that the drone can rely on fall-back systems and regain stable flight as soon as possible. In this talk, I will present novel methods to automatically recover and stabilize a quadrotor from any initial condition or execute emergency landing. On the one hand, this new technology will allow quadrotors to be launched by simply tossing them in the air, like a “baseball ball”. On the other hand, it will allow them to recover back into stable flight or land on a safe area after a system failure. Since this technology does not rely on any external infrastructure, such as GPS, it enables the safe use of drones in both indoor and outdoor environments. Thus, it can become relevant for commercial use of drones, such as parcel delivery.
Jean-Christophe ZUFFEREY (Sensefly)
Mapping and inspection drones for professionals
This short presentation will introduce senseFly’s professional drone systems. These include the eBee, the industry’s leading fixed-wing mapping drone. This safe, ultra-light aerial imaging tool can capture images spanning several square kilometres in a single autonomous flight. Loved by professionals around the world for its ease of use, the eBee is capable of operating in relatively strong winds and boasts unrivalled landing accuracy. senseFly’s eXom meanwhile is a next-generation quadrotor drone. Three years in development, it is the most sensor-rich civilian drone on the market, equipped with no less than seven cameras and five distance sensors, which work together to offer operators truly advanced situational awareness. This awareness and eXom’s bump-safe construction enable it to be used in close proximity of objects and structures, such as dams, bridges, and other industrial infrastructure.
Sebastien COURROUX (Novadem)
A new positioning technology for UAV allowing automatic inspection of structures without GPS coverage
The LPS technology of Novadem (Local Positioning System) allows for the first time to perform fully automatic inspection of structures. Based on its own terrestrial beacons network, the LPS technology reaches a centimeter-level accuracy while being independant from the satellite coverage. Its performances give new perspectives to inspections performed by UAVs thanks to flights that can be 100% automated even under structures. It is now possible to ensure the repeatability of the inspection over time. As a result, the security and efficiency of missions is significantly increased compared to a manual piloting. Add to that a more precise georeferencing of pictures allowing to optimise the flight plan on the areas to inspect and to rationalise the volume of data, thus simplifying the processing.
Tristan LOUIS (L'avion jaune)
From laser scanners to lidar systems dedicated to UAVs
The recent arrival of small laser scanners on the market has boosted the robotics industry, allowing to get a real time 3D map of robot’s surroundings. The UAV world has also shown its interest in such system for landing aids but also for surveying purpose.
If landing aids systems could be implemented without the need of an accurate absolute positioning, mapping system dedicated to aerial surveys must include high-end GNSS and INS sensors. Considerations about range and operational constraints such as weight and power requirement should also be taken into account.
After introducing different scanners available on the market, as well as considerations about the integration of GNSS and INS sensors, the YellowScan system will be presented. It results from 3 years of research and development for designing the lightest lidar system on the market. Data collected for different applications (topography, forest mapping, power line survey) will be presented. The surveys were performed using off the shelf multicopters as well as bigger unmanned systems. A discussion about data quality and the operational capacity of the different aircrafts in survey conditions will be held.
Antonio FRANCHI (LAAS-CNRS)
Observer-based Control of the Motion of Tethered Aerial Robots with Regulated Link Stress
The enormous popularity that aerial robots, commonly called UAVs (Unmanned Aerial Vehicles), are seeing in these days among the scientific community, but not only, is due to their versatility and applicability in a vast range of field, such as search and rescue, surveillance and patrolling, agriculture and civil monitoring and so on. Although their use as remote and unmanned sensors is dominant, they still present some limitations, as battery life and low payload that reduce their applicability where the task requires a very long or even constant gathering of measurements. In order to solve this problem, many recent works propose the tether solution. Indeed the use of a cable connecting the aerial vehicle to a fixed or mobile ground station can provide energy and an high-bandwidth communication link suitable for applications ranging from search and rescue to communication relay. The tethered flight solution could be exploited also to improve the hover stability in the presence of wind or during dangerous manoeuvres, e.g., landing and take off from a moving platform as a ship in the presence of rough sea.
In this talk I will briefly present the intrinsic properties of a tethered-aerial-robot system, such as exact linearizability with dynamic feedback, flatness, state observability, and trajectory feasibility. I will then present a new dynamic controller for the link stress and elevation using a nonlinear observer based on inertial-only measurements (accelerometer and gyroscope).
On the other hand, in the single robot case since the vehicle is constrained by the tether to fly around the fixed point, the working space, that is a simple circle, can result very limited for some applications. One solution is given by adding a second vehicle and a second link to the previous system.
In the second part of the talk I will then briefly present the investigation of a system composed by two aerial robots connected to each others and to a fixed point by two links, forming an actuated chain. The similarity with a planar two links robot immediately shows the advantages in terms of workspace, i.e., the second aerial robot in the chain (similarly to an end effector) can reach any point in the two dimensional working area, limited only by the full length of the system itself. Using this solution it is then possible to achieve the tracking of any desired Cartesian trajectory of the position of the second aerial robot while exactly controlling the stresses on the two links.
In this context I will briefly show a study on the controllability and observability of the system in the case of a set of plausible sensors, such as onboard accelerometers and encoders. This allows to close the control loop without the need of resorting to direct measures of the full state and of all the outputs to be tracked thus decreasing the cost of the solution in terms of sensorial equipment.
Augustin Manecy (Gipsa-lab & ISM, now at ONERA Toulouse)
A Bio-Inspired Visual Sensor and its Gaze Control Strategy for Robust Vision-Based Target Tracking by a Quadrotor
Insects, like hoverflies are able of outstanding performances to avoid obstacles, reject disturbances and hover or track a target with great accuracy. This suggests that refined mechanisms and treatments, even if they are minimalists, are perfectly optimized for the flight. The gaze orientation, stabilized by fast sensory motor reflexes, is probably one of the keys. Here, we proposed a new bio-inspired artificial eye of only 24 pixels, which mimics the main characteristics of the fly’s visual system, and which is able to locate a target at high refresh rates (up to 1 kHz). This eye was mounted onto a new open-source quadrotor, with an active decoupling between the eye and the body of the robot. The gaze orientation was stabilized by implementing fast bio-inspired reflexes. We demonstrated experimentally, that such a system can be used to achieve accurate positioning with respect to a target. Thanks to the gaze control strategy, the robot was able to reject efficiently high disturbances, to take-off and track faithfully a target and to land automatically onto a moving target.
Martial Sanfourche (ONERA Palaiseau)
Vision-based navigation for Micro Aerial Vehicles in complex
UAV are often presented as an economical means to acquire data for
structure inspection or environment monitoring. The current state of
technology allows to accomplish these missions in an automatized way in
some restrictive conditions like good reception of GPS signal or use of
an external multi-camera localization systems (like Vicon).
We present some research works conducted at ONERA about MAV autonomy in
GPS-denied indoor environments.
The proposed solution exploits the images acquired by low-cost 3D
sensors (stereo or kinect-like) and the measurements acquired by a low
grade IMU to estimate the vehicle trajectory at high frequency and to
model the environment. This model and the vehicule state are used as
inputs by a guidance and navigation system, able to avoid obstacles and
adapt the trajectory in order to explore autonomously an unknown
environment. This work has been rewarded with a second place (among 21
teams) during the simulation contest of the EUROC challenge “Plant
Servicing and Inspection.
Pascal Morin, Head of the RTE-UPMC Research Chair on “Autononous mini-drones”,
ISIR, UPMC, Paris