Self Driving With RL 

To enable the cars of the future.

Team Grasp

Autonomous driving has been widely hailed as a major solution to many transportation related issues such as road safety, congestion, cost of transportation and efficiency. While many selfdriving solutions are being developed today, they require long development efforts and do not scale well to new scenarios.

In this project, we aim to mitigate these challenges by learning to drive with reinforcement learning. This project aims at building an autonomous driving system capable of performing point-topoint navigation. This will be achieved with one or more reinforcement learning agents acting together to generate safe trajectories to the user-specific destinations. The agent shall perceive the world from a multi-modal sensor suite and high-definition map information, primarily from the simulation. Using the learned knowledge of how to drive, the agent will be able to perform the following tasks:

• Obey simplified traffic rules

• Perform point-to-point navigation

• Avoid obstacles along the planned path

• Follow a leading vehicle

• Change lanes when desired

• Negotiate with other agents at intersections

For further details look at our report at the link here.