rgbravi / autonomous-search-and-rescue-agent-using-deep-reinforcement-learning

Simulation of Autonomous search and rescue operation in a fire outbreak environment using Deep Reinforcement learning. (Project under development 💻)

Autonomous Search and Rescue Agent using Deep Reinforcement Learning

Domain: Artificial Intelligence

Problem Statement:

We are living in a world full of blooming autonomous technologies that can come as a great help for disaster management. It is highly risky for humans to enter risk prone areas like a fire outbreak in a building/any place, for rescuing those trapped inside, so it will be better to use self-trained agents(drones) to search and rescue the victims trapped inside such disaster-struck environments.

Abstract:

Rescue operation is highly risky for humans to enter risk prone areas like a fire outbreak in a building/any place, for rescuing those trapped inside. The existing protective gear used by the rescue team is fire proof but there is still room for risk due to falling debris in a collapsed building or suddenly collapsing walls and the sustainability of the rescuer is limited by the amount of the oxygen in the oxygen cylinder. Our idea is to simulate an environment wherein, an autonomous rescue agent finds its way through the disaster-inflicted environment, wading in through obstacles, broken walls, sudden flares of fire and locates the victims by means of available data (like sound waves of screams of people trapped inside, heat radiation etc). After locating the trapped victims, the agent tries to extinguish the fire around them using available resources and it sends a signal to the external rescue team (consisting of people) indicating the exact location of the victims. Likewise, it identifies all the clusters of victims trapped inside the building and does the same process for every cluster. The agent’s brain consists of a neural network that aids it in the decision-making process during its navigation. The agent is trained in a generalised sample environment using deep reinforcement learning. This project will eliminate the loss of lives of people in the rescue team and it will also help the rescue team to exactly locate the trapped victim inside the disaster struck building using the location information sent by the agent.

Dependecies:

The following dependencies need to be installed to run this project,

• Python
• Pygame
• Pytorch
• Tkinter

Snapshot of a successful trail:

Testing Environment snapshot: