Multi-Robot Mapping and Exploration of Environments with Hazards

Networks of robotic sensors have the potential to safely collect data over large-scale, unknown environments. They can be especially useful in situations where the environment is unsafe for humans to explore. In many such situations, robots are also susceptible to hazards. It is important to design exploration and mapping algorithms that are hazard-aware, so that the robotic sensor network can effectively carry out its task while minimizing the impact of individual robot failures. In this work we describe an algorithm, based on an analytic expression for the gradient of mutual information, that enables a robotic sensor network to estimate a map of events in the environment while avoiding failures due to unknown hazards.