Backup Warning and Recovery System on Autonomous Vehicles

Autonomous vehicles are more and more popular but they are vulnerable. People cannot fully trust the vehicles so a backup system should be deployed on the cloud.

What it does

Simulators of autonomous driving has been developed to test 2 kinds of attack: zero-alarm attack and hidden attack, we use CUSUM statistics to detect the time-veries signal change. A safety deadline was given by reachability analysis based on the system model. A seq2seq RNN was trained to recover the spoofed speed of the car. Depolyed Scripts to GCP.

How we built it

The scripts are fully developed by python through Jupyter Notebook And a computing engine on GCP should be responsible for running them on cloud.

Challenges we ran into

Hard to simulate non-linear vehicles, so a linear model was applied. The trainning takes time, and the parameters tuning is a pain. Not familiar with GCP.

Accomplishments that we're proud of

The system can make autonomous driving much more safe, since the cloud servics is much safer than the system on a car. The backup recovery may become the next generation autonomous driving techs.

What we learned

Try to implement something fast and efficient by existed repo Don't make the system too complex Try to divide the works in acceptable components

What's next for Backup Warning and Recovery System on Autonomous Vehicles

Make it to a conference paper by the end of the semester. Apply it to complex non-linear models. C-version of codes to speed up the scripts.

Built With

• control
• keras
• python
• reachability
• simulation
• tensorflow