SPEAR

The popular unmanned aerial robot designs are inspired by manned aerial vehicles from the early 20th century. Since the 2000s, quadrotor drones have dominated the small aerial robots domain due to their structural simplicity and agility. Is further development bound to incremental improvements in their design alongside progress in the autonomy stack? We argue that future research cannot focus on incremental or separate steps in drone design and autonomy alone. Despite the outstanding progress in various domains –from control to perception and beyond - conventional multirotor systems are subject to multiple limitations that constrain their utilization envelope. 

Furthermore, today's designers often need to tailor their robots to a specific task in a particular application domain, which however is particularly time-consuming. Unlike the current practice, we propose a novel approach to change the paradigm in the design process. We depart from the compartmentalized approach of human-engineered designs and investigate the holistic co-synthesis of task-specific flying robot embodiment and autonomy through a synergistic combination of evolutionary algorithms and deep learning for navigation. 

We aim to show the benefits of breeding unconventional ""bodies"" and ""brains”. Bodies rely on an evolutionary combination of rotary wing components, and soft and rigid elements, whereas brains exploit the latest progress in deep neural networks. This fundamental change in the design procedure offers a unique pathway towards more capable, more resilient, intrinsically safe flying robots. Upon its success, SPEAR will drive the robotics community forward and towards automatically designed and task-optimized flying machines with superior performance.