state initialization in each RL step #84

Hello,

In the paper it is mentioned:

During each RL step: (1) The robot is initialized at the real-world state st_r.

However, in the open-source code I only found that the torque is computed using the real-world motion action + delta action. I did not see where the robot’s state is actually overwritten/initialized with the real-world state.

Could you please further explain how to understand the statement that at each RL step the robot’s state is reset to the real-world state? I might have misunderstood the implementation.

Thank you! 🙏