Sushant Suryawanshi

• Global Planner :-

- Designed and implemented a global planner utilizing a Hybrid A* algorithm to facilitate vehicle motion within a custom scenario. This involved integrating static actor vehicles generated within Unreal Engine and leveraging the Automated Driving Toolbox from Matlab.

- Implemented Voronoi field diagram for restructuring the obstacle free space for planning in Matlab.

• Perceiving surrounding :-

- Implemented obstacle detection capabilities for the ego vehicle amidst dynamic obstacles using the Automated Driving Toolbox in Matlab.

- Additionally, collected relative distance data over time for various obstacles.

• DRL based local planner :-

- Developed a deep reinforcement learning-based local planner capable of navigating dynamic actor vehicles in a highway scenario.

- Implemented actor-critic functionality using the DDPG algorithm within the reinforcement learning framework.

- Demonstrated ~15% enhancement in safety compared to traditional optimization-based planners.

• Imitation Learning :-

- Currently engaged in implementing Behavior Cloning as the initial phase of Imitation Learning to emulate the behavior of Model Predictive Control (MPC) for generating velocity profiles.

- Utilizing raytune for hyperparameter tuning in supervised learning processes.

- Developing an Imitation Learning pipeline incorporating the Dagger algorithm to enhance network robustness, particularly in critical states.