An online evolving method named evolving Finite State Machine (e-FSM) can determine unknown states (situations) and identify transitions. At the moment, it is similar to Markov Chain, but its structure evolves over time. This approach enables controllers to recognize unexpected situations and learn optimal decisions over time. Also, the e-FSM is fully explainable, while Deep Neural Network is not.
Driving intention is recognized by determining and estimating a driving characteristic, and lane-change behaviors are predicted by training a neural-network based predictor