Research

In our group, we are focusing on developing state-of-art methods for multi-agent control and coordination using Reinfrocement Learning algorithms. Multi-agent system is composed of multiple individual agents which could be homogeneous or heterogenous. Each agent works to achieve its local objective in a manner such that the whole system could achieve a global objective. There are various challenges in multi-agent setup from the perpective of reinforcement learning such as non-stationarity of the environment, huge joint state-space. etc. Our group is focusing on using the recent advancements in Deep Reinforcement Learning to create efficient and robust Multi-agent Reinforcement Learning Models.

As an application domain, we are extensively looking into the usage of multi-UAV system for providing solutions in various fields such as - Help & Rescue during natural calamities, Surveying and Tracking of large sites, Indoor localization and Map construction, payload delivery in urban landscape, etc. To achieve these application specific objecitves, we employ domain specific contraints along with off-the-self computer vision tools. Initial algorithms development and rigourous testing of different parameters are performed using simulations. Aftewards, we strive to create a proof-of-concept with our available resources.