* Responsible for data-driven and learning based behavior planning algorithms for highway scenarios.

* Developed a recurrent network based multi-task classification and regression model for specific highway behavior generation.

* Deployed model through a C++ module for both simulation and vehicle embedded platforms using a wrapped Nvidia TensorRT engine.

* Developed a data pipeline for data processing from simulations and vehicle. Pipeline involved pre-processing data for consumption by model, as well as reporting statistics to detect anomalies with dataset (e.g. Class imbalance)

Designed and Developed longitudinal vehicle dynamics models for Software in the Loop Simulation and Testing of hybrid vehicles using sparse temporal data based System Identification techniques. Achieved up to 97% model fit accuracy for lateral models using only a 5 minute section of data.

Achieved up to 74% improvement in lateral and heading error tracking using augmented LQR controller for lateral dynamics. Improvements tested and verified in simulation and vehicle tests based on Baidu - Apollo AD architecture.