To Start Training the model, First load the robot in gazebo using l2.py launch file, then start com.py for the current and next state data, then run the appropriate RL script.
Figure 1: Control Architecture
Figure 2: Ros2-Gazebo Communication Structure
This project tackles the complex task of controlling a quadruped robot using model-free reinforcement learning (MFRL) within the ROS2 and Gazebo simulation environment. Quadruped robots are versatile but face challenges in adapting to dynamic environments as well as require complex control de- sign to operate robustly in dynamic environments. The integration of ROS2 and Gazebo is crucial, serving as a bridge between simulation and real-world deployment.Our solution combines MFRL with ROS2 and Gazebo, providing an efficient platform for testing and transitioning learned policies to physical robots. The ROS2 framework ensures streamlined communication, while Gazebo offers a realistic simulation environment. This approach enhances adaptability and autonomy, crucial for applications like search and rescue.The significance lies in mitigating risks associated with direct experimentation on physical robots. By allowing robots to learn and optimize control policies in simulation, as real hardware is susceptible to damage when training the model in a real robot. This project explores an alternative solution for quadruped robot control, addressing challenges posed by real-world complexities and advancing autonomous robotic systems.