Risk-Aware Conditional Planning with Applications in Autonomous Vehicles

  • Rashid Sulaiman Alyassi

Student thesis: Master's Thesis


According to the World Health Organization, around 1.25 million people globally die from car accidents annually. Human controlled vehicles introduce the risk of failure due to various factors related to intrinsic human limitations. A solution comes in the form of vehicle automation. While autonomous vehicles have been showcased with a high degree ofautonomy, theystillpresentariskfactorwithmultiplerecordedcollisionsandfatalities. To alleviate that, one has to implement a risk-aware decision pipeline for the autonomous vehicle system that can capture and propagate risk properly. Vehicle automation is divided into four subsystems: sensors, perception, planning, and control. In this research, our focus is on the planning subsystem, which is a central component for safe operation. The planning subsystem consists of route planning, intent prediction, behavior planning, and trajectory planning. Our main contributions are: a polynomial-time approximate solver for a risk-aware behavior planner, a highly accurate (99%) intent prediction model for the roundabout scenario, and a quick (sub 50µs) risk estimation functions required for trajectory planning. Furthermore, we provide experiments that represent how the provided approach operates together by constructing a smart intersection for human-driven and autonomous vehicles utilizing Vehicle-to-Infrastructure (V2I) communication. Existing risk-aware partially observable systems for autonomous vehicles suffer from running timeissues. Thecentraltheseofthisresearchistoprovidefastermethodsforarisk-aware, partially observable autonomous vehicle system capable of running in an online fashion.
Date of AwardOct 2020
Original languageAmerican English


  • Autonomous Vehicles
  • Robotic Planning
  • MDP
  • Intent Prediction

Cite this