Browsing by Subject "partially observable Markov decision process"

Sort by: Order: Results:

Now showing items 1-1 of 1
  • Rantapelkonen, Antti (Helsingin yliopisto, 2020)
    A self-driving car must be able to observe and predict behavior of other road users in the environment where the states are only partly observable. Sensor data provides accurate identification of type, location, speed, and orientation of other road users, but predicting their intentions is difficult for artificial intelligence. The problem can be solved with partially observable Markov decision process (POMDP), which provides mathematical framework for decision making in uncertain situations. Nevertheless, the challenge for POMDP is real-time computation. Solving POMDP is mathematically intractable, therefore, POMDP solvers are used for approximations that are sufficiently accurate. Additionally, scalability for adequate number of other road users is challenging for many POMDP solvers. This master thesis is a literature survey in which four research papers are analyzed. The research papers provide solution for uncertainty in self driving cars decision making using POMDP with different solver algorithms in intersection and crosswalk scenarios.