Visual Navigation via Reinforcement Learning Rewards in CARLA
Research Summary
Research Summary
In this project, I experimented with various reinforcement learning (RL) algorithms in the CARLA simulator. The goal is to create RL agents that can learn different driving policies using only visual (RGB) input.
For instance, the following video shows how a DQN agent learns to make a right turn over time.
The agent can also learn to avoid obstacles (parked cars) in the scene.
What about turning and avoiding obstacles in a roundabout scene?
Bird's-eye View
First-person View