Building environments with typical human social activities that embed intent communication in their hidden representation, e.g handling an item to another agent, waving/pointing in one direction or another, or even a more implicit representation as instantaneous motion acceleration/deceleration, helps reinforcing the learning model of an autonomous agent towards an optimal decision-making given unpredictable environments like warehouses and hospitals. The student will work on building such environments with the goal of encoding them in a decision making model that learns how to respond to unpredictable intent communication. Environments will be built in simulation (mainly Gazebo) using ROS actors modeling and programming will be in C++ (mainly) /python.
Baseline of work:
(a) crowd and groups modeling: https://github.com/srl-freiburg/pedsim_ros
(b) closed human actors simulation/modeling: https://classic.gazebosim.org/tutorials?tut=actor&cat=build_robot