Towards an Experimental Autonomous Blimp Platform.

Abstract

Abstract — In this paper, we present the design of an autonomous indoor blimp. We describe the individual low-weight components of an embedded system including actors and sensors which achieve a total weight of less than 200 grams. Both, the modular hardware components and the use of generic hardware interfaces facilitate the adaption to different actor and sensor systems. Similar to the flexibility of the hardware components, the user benefits from a generic software framework. Here, an onboard Linux operating system and device driver interface ease the implementation of robotic applications and control tasks. The main challenge of designing a small blimp in regard to autonomous operation is the efficient interplay of the individual software and hardware components. In this work, we show approaches to achieve this and to obtain a system which can easily be applied to other robotic systems with constraints in weight and available space. We evaluate the performance of the components and demonstrate their integration in a reinforcement learning setting. Index Terms — reinforcement learning and navigation I.