The lack of extensive, labelled data sets representative of dynamic multi-agent environments currently hinders our ability to use machine learning techniques to develop autonomous driving systems. In order to conduct experiments on alternative algorithms for real-time collaborative sensing, there is a need for programmable and repeatable dynamic scenarios.

This project provides a scaled-down, physical implementation of a programmable environment simulator for reliably gathering data and testing the effectiveness of collaborative sensing algorithms in communication constrained environments. Researchers will be able to evaluate, train, and test potential solutions in real time with the ability to simulate different traffic scenarios as well as different sensing environments. If user requires more variety, the design allows for some reconfiguration and additions to the field. In total, this project aims to overcome some of the issues with testing machine learning and collaborative sensing algorithms.