Autonomous Car

The goal of this effort was to develop a test bed that would facilitate research on various aspects of autonomous cars including machine learning, sensors, and cybersecurity. The test bed includes first-person driving controls that may be used to generate data sets for training of the machine learning algorithms that will recognize obstacles and drive the smart car. We currently have the ability to inject an array of simulated cyber attacks such as brake failure, stuck accelerator, and delayed response. This capability allows us to observe the response of the driving algorithm in real-time as the system experiences the attack.

Below are photos of the autonomous car test bed running the OpenDS driving simulation software.

Click Photo Above to Enlarge

Not pictured are servers that include a variety of NVIDIA GPUs (P100, K20) that are used for for training of neural network classifiers, a set of high-resolution cameras capable of capturing training images from real vehicles without scan-induced distortions, and a set of NVIDIA single board computers (TK1, TX1, and TX2 boards) that can host the trained machine learning algorithms as they control cyber physical systems.

Below is a photo of the test bed which is mounted on a mobile platform to allow transport to STEM outreach activities.

Click Photo Above to Enlarge

Development work is on-going so please check back later for additional updates.