Non-homogeneous autonomous robots for search and rescue in the aftermath of disasters

Autonomous search and rescue robots can save the lives of search and rescue staff members and the trapped victims. Ideally, search and rescue robots should make quick decisions that are accurate despite of the uncertainties in the search and rescue environment, and are optimal in terms of the time, safety, and coverage of the area.

When formulated accurately, the optimal control problem of search and rescue robots is computationally intractable online. The main issues are nonlinearities, large size of the optimization time window resulting in large number of optimization variables, and uncertainties due to the unknown nature of search and rescue. To tackle these challenges, currently heuristic or simplified optimization-based approaches are used, which lack optimality guarantees and accuracy in the provided decisions. Therefore, human rescuers should still interfere in mapping the search and rescue area, which may be the most life-threatening stage of search and rescue due to the unknown risks.

To resolve these fundamental barriers, in this project we introduce novel mathematical reformulations of the control optimization problems of search and rescue robots. These new formulations provide balanced trade-offs between computational complexity and solution accuracy. Furthermore, we will develop new concepts that integrate AI and optimization-based approaches and systematically incorporate uncertainties in the control optimization problems and resulting decisions.

Finally, by proposing a flocking unit of non-homogeneous robots, ranging from computationally advanced robots to microscopic simple drones, we implement the developed control and decision making approaches to simulated search-and-rescue scenarios.