Flying robots have the unique advantage of being able to move through the air unaffected by the obstacles or precipices below them. This ability quickly becomes a disadvantage, however, as the amount of free space is reduced and the risk of collisions increases. Their sensitivity to any contact with the environment have kept them from venturing beyond large open spaces and obstacle-free skies. Recent efforts have concentrated on improving obstacle detection and avoidance strategies, modeling the environment and intelligent planning to navigate ever tighter spaces while remaining airborne. Though this strategy is yielding impressive and improving results, it is limited by the quality of the information that can be provided by on-board sensors. As evidenced by insects that collide with windows, there will always be situations in which sensors fail and a flying platform will collide with the obstacles around it. It is this fact that inspired the topic of this thesis: enabling flying platforms to survive and recover from contact with their environment through intelligent mechanical design. There are three main challenges tackled in this thesis: robustness to contact, self-recovery and integration into flight systems. Robustness to contact involves the protection of fast-spinning propellers, the stiff inner frame of a flying robot and its embedded sensors from damage through the elastic absorption of collision energy. A method is presented for designing protective structures that transfer the lowest possible amount of force to the platform's frame while simultaneously minimizing weight and thus their effect on flight performance. The method is first used to design a teardrop-shaped spring configuration for absorbing head-on collisions typically experienced by winged platforms. The design is implemented on a flying platform that can survive drops from a height of 2 m. A second design is then presented, this time using springs in a tetrahedral configuration that absorb energy through buckling. When embedded into a hovering platform the tetrahedral protective mechanisms are able to absorb dozens of high-speed collisions while significantly reducing the forces on the platforms frame compared to foam-based protection typically used on other platforms. Surviving a collision is only half of the equation and is only useful if a flying platform can subsequently return to flight without requiring human intervention, a process called self-recovery. The theory behind self-recovery as it applies to many types of flying platforms is first presented, followed by a method for designing and optimizing different types of self-recovery mechanisms. A gravity-based mechanism is implemented on an ultra-light (20.5 g) wing-based platform whose morphology and centre of gravity are optimized to always land on its side after a collision, ready to take off again. Such a mechanism, however, is limited to surfaces that are flat and obstacle-free and requires clear space in front of the platform to return to the air. A second, leg-based self-recovery mechanism is thus designed and integrated into a second hovering platform, allowing it to upright into a vertical takeoff position. The mechanism is successful in returning the platform to the air in a variety of complex environments, including sloped surfaces, corners and surface textures ranging from smooth hardwood to gravel and rocks. In a final chapter collision energy absorption and self-recovery mechanisms are integrated into a single hovering platform, the first example of a flying robot capable of crashing into obstacles, falling to the ground, uprighting and returning to the air, all without human intervention. These abilities are first demonstrated through a contact-based random search behaviour in which the platform explores a small enclosed room in complete darkness. After each collision with a wall the platform falls to the ground, recovers and then continues exploring. In a second experiment the platform is programmed with a basic phototaxis behaviour. Using only four photodiodes that provide a rough idea of the bearing to a source of light the platform is able to consistently cross a 13x2.2mcorridor and traverse a doorway without using any obstacle avoidance, modeling or planning.