Why does it matter?
Current drones have limited decision making whenever they have incomplete information about the environment they need to interact. Under the context of Search and Rescue (SAR), surveying environments such as collapsed buildings, bushland and rainforests are often difficult to access or lack a detailed map. Onboard drone identification and localisation of victims is challenging due to partial (or full) occlusion between the victims and surrounding objects, possible low lighting conditions, noise in the data, and low image resolution.
This project aims to investigate the incorporation of a sequential decision-making model framework for fully autonomous UAV operations, able to navigate under unstructured environments and reduce levels of target detection uncertainty. The design of the UAV framework and system architecture should allow onboard execution of computer vision and decision-making algorithms in resource-constrained hardware, removing the dependency of the UAV on external ground control stations and communication systems, so it can interact with the environment by itself and accomplish the flight mission unattended.
Real World Impact
Disaster management and rescue squads require fast and reliable information to prioritise emergency response in affected zones. It is therefore important to design drone systems with autonomous decision-making which can quickly localise, identify and quantify potential victims. The majority, if not all, of real-world environments are unstructured. Enabling autonomous decision-making in unknown environments will bring a new range of applications in which drones are currently limited to provide support.
Other Team Members
- Dr Peter Caccetta
- Dr Conrad Sanderson