Webpages tagged with «reinforcement learning»

Guide ground investigation (GI) with reinforcement learning

Information from GI-drilling is used to build a ground model with layers of clay, gravel, sand, etc., and bedrock surface. The ground model is used for designing buildings, infrastructure and mapping natural hazard risks, such as a quick clay landslide. The agent should minimize the necessary holes to build a ground model for a defined area, with sufficient accuracy, and detect all risk zones.

This is an applied problem where you can contribute to safer, more sustainable, and more efficient construction of buildings and infrastructure + basis for natural hazard assessment. The problem is yet to be sufficiently solved. You will test your model (trained RL-policy) on real ground investigation data from the Norwegian Geotechnical Institute (NGI).

Different types of drones are used for a range of environmental monitoring activities. The hardware involved is still quite costly to acquire and deploy, so various ways to make them more cost-effective is an active area of research. One way to increase the usefulness of autonomous robots, is enabling them to perform more complex tasks, or multiple tasks. In this project, you will explore reinforcement learning (RL) with the aim of training a drone to detect and map plumes of gas emission.