There are not yet autonomous flying robots capable of manoeuvring in small cluttered environments as insects do. Encouraged by this observation, this thesis presents the development of ultra-light flying robots and control systems going one step toward fully autonomous indoor aerial navigation. The substantial weight and energy constraints imposed by this indoor flying robots preclude the use of powerful processors and active distance sensors. Moreover, flying systems require fast sensory-motor mapping despite their limited processing power. In order to cope with those apparently contradictory constraints, our approach takes inspiration from flying insects, which display efficient flight control capability in complex environments in spite of their limited weight and relatively tiny brain. In particular, they are able to stabilise their course, avoid obstacles and control their altitude, which represents the basic mechanisms we want to have on an indoor flying robot. To achieve efficient flight control, insects rely essentially on two sensory modalities: vision and gyroscope. They possess two low-resolution, compound eyes which are particularly sensitive to image motion (optic flow). In their visual system, some neurons are known to be responsible for detecting self-motion and impending collisions based on optic-flow. Gyroscopic information coming from two mechanosensors located behind the wings complements visual cues in tasks such as gaze and course stabilisation. In this thesis, we explore the application of such biological principles to develop navigation controllers for indoor flying robots. In particular, we address the problem of how low-resolution vision and gyroscopic information can be mapped into actuator commands in real-time to maintain altitude, stabilise the course and avoid obstacles. As an alternative to hand-crafting control systems based on biological principles, in a second phase, we take inspiration from the evolutionary process that eventually generated those animals and apply artificial evolution to search for alternative control systems and behaviours that can fit the constraints of indoor flying robots. Instead of replicating the biomechanics of insect flight, our targeted robotic platform is a fixed-wing airplane capable of flying indoors at very low speed (<1.5m/s). This testbed weights only 30-grams and is equipped with several miniature cameras and a small gyroscope. In order to progress gradually in the task of automating indoor flying robots, two other platforms have been developed, namely a miniature wheeled robot and a small indoor airship. All three robotic platforms feature very similar sensors and electronics in order to facilitate the transfer of software modules and control strategies. Applying the proposed bio-inspired approach, we succeeded in automating the steering (course stabilisation and obstacle avoidance) of the 30-gram airplane in a square textured arena. Then, using artificial evolution with the airship, we obtained alternative navigation strategies based on the same sensory modalities.