Intelligent Self-driving System (ISS) is a modular framework written in Python and C++ with the aim to build an extensible workspace tailored to research. This framework will contain both traditional and deep learning algorithms for self-driving related tasks such as perception, localization, mapping, prediction, planning, and control. The modular design with minimal dependency on external libraries can provide a transparent and clean workspace for researchers to evaluate algorithms for autonomous driving systems.

ISS Architecture

ISS Architecture

Simulator Demos

At present, the ISS framework has the capability to deploy and test algorithms using data generated by a simulator. Upon integrating sensor data from the CARLA simulator into our framework, we can evaluate a range of algorithms. Additionally, corresponding control algorithms can be employed to maneuver the simulated vehicles. Here are some demonstrations.

A real-time demonstration of the PointPillar 3D detection algorithm within the Carla simulator

A demo showing the local behavior planning

Minicar Demos

In addition to simulation results, if the sensor data from a real minicar is transmitted back to ISS via ROS, the ISS framework can leverage the sensor data to accomplish a variety of tasks and exert control over the physical minicar. We present below three videos showing the interaction of ISS with real minicars.

Two demos featuring physical minicars navigating indoor and outdoor environments, respectively

A demo showcasing the A-LOAM SLAM algorithm using data from physical minicars