Decision Making for Autonomous Vehicles at Signalized Intersection under Uncertain Traffic Signal Phase and Timing Information

In real environments, autonomous vehicles must be able to deal with uncertainties related to the measurements provided by their perception system. Not taking such perception uncertainties into account can lead the vehicle to take erroneous decisions and cause accidents. Excessive speed rate change variations and red light crossing at signalized intersections are special cases of this problem. This paper presents a decision making for autonomous vehicles that considers uncertainty in timing information, but also in traffic light color (signal phase) measurement at signalized intersections. A Partially Observable Markov Decision Problem (POMDP) model is proposed in order to deal with the problem of partially observability of both signal phase and timing. By incorporating the perception system uncertainty, the POMDP model can reliably predict signal phase transitions, avoiding reacting in a too reactive manner and running red lights. Results show that the proposed POMDP model is able to estimate the true values of the signal phase and timing as more observations are gathered, which allows better decisions when compared to deterministic approaches.