How to deploy trained policys

[Jak1283]

Hello,
I'm working on training a custom reinforcement learning policy for the Unitree G1 robot's navigation task, where the robot learns to navigate to a specific (x, y) location using UniformPose2dCommand. This is based on the pre-existing Unitree-G1-29dof-Velocity policy as a starting point (PreTrainedPolicyActionCfg).

Training and export work fine—I get a policy.pt file via play.py. However, I'm stuck on deployment:

• To MuJoCo simulation: I can't get it running with g1_ctrl after modifying main.cpp.
• To real robot: No clear guide for integrating custom policies beyond the defaults.

A step-by-step guide (or example code/YAML snippets) for deploying self-trained policies would be incredibly helpful, especially for custom tasks like navigation.

Error: Trying to deploy the custom policy via g1_ctrl

terminate called after throwing an instance of 'YAML::TypedBadConversion<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >'
  what():  bad conversion

I believe a new class (for the custom policy loader or navigation handler) needs to be created and added to the build, but the question is where (in which header/source file) and how (CMake integration, inheritance from existing policy class).

[Agnel-Wang]

You can use gdb or VS Code Debug to check exactly where the YAML import is failing.

[bangboomtata]

Hello, I'm working on training a custom reinforcement learning policy for the Unitree G1 robot's navigation task, where the robot learns to navigate to a specific (x, y) location using UniformPose2dCommand. This is based on the pre-existing Unitree-G1-29dof-Velocity policy as a starting point (PreTrainedPolicyActionCfg).

Training and export work fine—I get a policy.pt file via play.py. However, I'm stuck on deployment:

• To MuJoCo simulation: I can't get it running with g1_ctrl after modifying main.cpp.
• To real robot: No clear guide for integrating custom policies beyond the defaults.

A step-by-step guide (or example code/YAML snippets) for deploying self-trained policies would be incredibly helpful, especially for custom tasks like navigation.

Error: Trying to deploy the custom policy via g1_ctrl

terminate called after throwing an instance of 'YAML::TypedBadConversion<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >'
  what():  bad conversion

I believe a new class (for the custom policy loader or navigation handler) needs to be created and added to the build, but the question is where (in which header/source file) and how (CMake integration, inheritance from existing policy class).

did you manage to find where to create and add a new class (for the custom policy loader or navigation handler)