Michelle-NYX's repositories
traffic-simulator-Q-learning
We propose a driver modeling process and its evaluation results of an intelligent autonomous driving policy, which is obtained through reinforcement learning techniques. Assuming a MDP decision making model, Q-learning method is applied to simple but descriptive state and action spaces, so that a policy is developed within limited computational load. The driver could perform reasonable maneuvers, like acceleration, deceleration or lane-changes, under usual traffic conditions on a multi-lane highway. A traffic simulator is also construed to evaluate a given policy in terms of collision rate, average travelling speed, and lane change times. Results show the policy gets well trained under reasonable time periods, where the driver acts interactively in the stochastic traffic environment, demonstrating low collision rate and obtaining higher travelling speed than the average of the environment. Sample traffic simulation videos are postedsit on YouTube.
DreamNet
CS230 Project: In this project, we investigate and evaluate the performance of the state-of-the-art model for instance segmentation, Mask R-CNN, on the newly-released Mapillary dataset, whose images focus specifically on driving scenes. We transfer the learning results from the pre-trained weights, fine tune the final layers for Mapillary Datasets. The result shows a significant improvement in precision measurements from the baseline, and achieves at a surpassing performance than benchmarks.
CollisionAvoidance
forked from https://github.com/StanfordASL/Pigeon.jl.git