Research Projects

Forest fires are one of the most common and serious emergencies. They disturb the natural areas resulting in important losses of plant and animal diversity, and also urban areas resulting in human and material losses. The detection and response time of mitigating forest fires is very important to reduce the damage. Himachal Pradesh is one such state in India which experiences several forest fires naturally or by human mistakes. There is a serious need to bring a change in the conventional way of mitigating these wildfires. With the implementation of robotic technology like a swarm of drones, forest fires can be curbed in a safer and faster way. But the reliability of the drones for implementation in such scenarios is very much important. As they may be working along with humans and any undesired action from the swarm may lead to additional damage besides the forest fires. So, the algorithms for handling the anomalies should be concrete and flexible to adapt to real-time chaotic situations. This project proposes an experimental framework to handle such faulty situations when drones are working collaboratively for mitigating forest fires. To effectively handle the firefighting task, the formation should be altered during the run time according to the spread of the wildfire. So a time-varying formation tracking algorithm (TVFT) is proposed here where an arbitrary formation can be decided online. Any anomaly in the swarm is diagnosed and detected online by the design of a distributed observer at each drone. The isolation of the faulty agent/reconfiguration of the swarm is decided based on fault magnitude. The faulty agent with at least one working rotor, if isolated, is given proper control inputs such that it makes a safe landing. The project is focussing towards a practical approach where only the input and output data is available at a given time. For the fault-tolerant algorithms to be implemented effectively the knowledge of the faulty model is important. Motivated by the recent successful implementation of a data-driven approach “Koopman framework” which gives the global linear representation of the nonlinear model in a pure data-driven way, the proposed model of fault tolerant control of a swarm of drones will be investigated with the Koopman framework as the core idea of the work. With the Koopman framework, the algorithm of TVFT, FTC and reformation of the swarm by isolation of faulty agent, and safe landing of faulty agent will be implemented on a swarm of crazyflies (miniature programmable drones). This work will result in an experimental framework towards the mitigation of forest fires which can further be developed into real-time solutions. It strengthens and supports the implementation of a swarm of drones in many real-time applications. Also, it encourages the researchers to explore the Koopman framework and its idea while working with nonlinear control systems.