Year: 2021
Type: dissertation
Abstract: The dynamic nature of vehicles and pedestrians in urban environments poses a challenge for autonomous driving to safely make control decisions. We propose a reinforcement learning based motion-planning algorithm for the autonomous vehicle to interact with a partially observable environment where the states will be obtained by LSTM, to enable the autonomous vehicle’s ability to impute information from the environment with no direct sensing method. To verify this algorithm, we conduct parametric study and check the collision rate and time-to-complete (TTC), signifying the autonomous vehicle safely reaching the goal position without collision.
Author: Zhaoux Deng
Institution: University of Illinois Urbana-Champaign
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3187091474
