Deep Reinforcement Learning (PPO) in Autonomous Driving (Carla) [from scratch]

It is highly evident that autonomous vehicles will be the future and it will be a prominent vehicle category in the next decade. For this to be a success, the vehicle should be safe, reliable and provide a comfortable user experience. Autonomous driving must have sophisticated negotiating skills while taking right, left turns and while pushing ahead in urban areas. Reinforcement learning is considered as the main domain for learning driving policy. We propose a reinforcement learning approach using deep Q-learning approach which will extract the maximum reward from a large state space. We use CARLA, an open-source simulator for autonomous driving research. The outcome of this experiment is to resemble a real-life environment where the agent tries to overcome the obstacles using the data from the virtual sensors attached to the agent.

The code has been implemented in Carla Simulator with the help of Double DQN to train an agent how to drive autonomously using different architecture. To train the agent in command line pass the parameter "Train" and after training to test how the agent is behaving pass "Run" as parameter

No description available

Autonomous driving agent trained using a custom PyTorch CNN architecture for reinforcement learning with Deep Deterministic Policy Gradient (DDPG) in the CARLA simulator.

No description available

Implementing Autonomous Driving using Deep Reinforcement Learning in CARLA simulator

Research-oriented deep reinforcement learning project for safe autonomous driving in the CARLA simulator, using transformer-based perception and safety-aware decision making.