The AirBurr: A Flying Robot That Can Exploit Collisions Adrien Briod*, Adam Klaptocz*, Jean-Christophe Zufferey and Dario Floreano Abstract—Research made over the past decade shows the use of increasingly complex methods and heavy platforms to achieve autonomous flight in cluttered environments. However, efficient behaviors can be found in nature where limited sensing is used, such as in insects progressing toward a light at night. Interestingly, their success is based on their ability to recover from the numerous collisions happening along their imperfect flight path. The goal of the AirBurr project is to take inspiration from these insects and develop a new class of flying robots that can recover from collisions and even exploit them. Such robots are designed to be robust to crashes and can take-off again without human intervention. They navigate in a reactive way, bump into obstacles, and unlike conventional approaches, they don’t need heavy modeling in order to fly autonomously. We believe that this new paradigm will bring flying robots out of the laboratory and allow them to tackle unstructured, cluttered environments. This paper aims at presenting the vision of the AirBurr project, as well as the latest results in the design of a platform capable of sustaining collisions and self-recovering after crashes. Index Terms—Robust bio-inspired indoor flying robot. I. INTRODUCTION Flying robots have unique advantages in the exploration and surveillance of indoor environments presenting dangers to humans, such as caves, semi-collapsed buildings or ra- dioactive areas. Flight as indoor locomotion is interesting because it is not constrained by the morphology of the ground and can be used to navigate over obstacles more efficiently than ground-based locomotion. Current flying systems how- ever have difficulty in dealing with the large amount of obstacles inherent to such unknown environments. Collisions with this ’clutter’ generally result in crashes from which the platform can no longer recover. Many researchers thus focus on obstacle detection (using mechanisms ranging from optic flow [1], IR range sensors [2] or lasers [3]) and try to avoid collisions at all costs. However, the lack of global positioning (like GPS) and the unstable nature of flying platforms render this task increasingly difficult as the complexity of the environment increases, requiring advanced sensors, powerful processors and extensive modeling of the environment. As an example, All authors are with the Laboratory of Intelligent Systems, Ecole Polytech- nique F´ ed´ erale de Lausanne, 1015 Lausanne, Switzerland. Contact e-mail: [adrien.briod|adam.klaptocz]@epfl.ch. This work was supported by Armasuisse, competence sector Science + Technology for the Swiss Federal Department of Defense, Civil Protection and Sports. This research was supported by the Swiss National Science Foundation through the National Centre of Competence in Research Robotics. *Adrien Briod and Adam Klaptocz contributed equally to this work. Fig. 1. The AirBurr robot (depicted in the above artist’s impression) will be able to explore cluttered indoor environments autonomously. A contact- sensitive structure allows the robot to navigate around obstacles by flying away from them once it touches them. During collisions, the structure protects the robot from damage. If the robot falls to the ground, it can actively upright itself thanks to its legs and take off again without human intervention. the most advanced and successful method to date is the simultaneous localization and mapping (SLAM) approach, which allows absolute positioning in a map built by the system itself using high-precision on-board sensors. SLAM- enabled platforms in the 1-2kg weight range equipped with laser scanners or cameras and relatively powerful processors have realized very successful demonstrations in unknown indoor environments [4], [5], [6]. However, such platforms are prone to catastrophic mission-ending crashes if a collision happens with an obstacle that failed to be detected by the sensors. Indeed, due to their weight, such platforms are relatively fragile and cannot afford making contact with the environment, and thus far have only been demonstrated in fairly structured environments that only contain rooms, hallways and openings to simulate windows. We aim at taking a different approach to tackle indoor autonomous flight. Instead of using heavy sensing, modeling and control, we take inspiration from nature’s most success- ful flyers such as insects that are capable of impressively dynamic flight indoors using only local information and simple navigation algorithms. The main source of inspiration however is how insects react when their algorithms fail. Though they often crash into transparent windows or low- contrast walls, their flexible bodies absorb the impact energy