This project is based on the Robopianist environment. While the Robopianist environment focuses entirely on simulation, this project implements a sim2real2sim transfer. You can find videos of this project and further details at lasr.org.
The project contains multiple entry points for different use cases.
| File | Purpose |
|---|---|
train.py |
Starts a single training training. |
training_scheduler.py |
Schedules multiple training runs with varying parameters. |
robopianist/execution/policy_executors.py |
Executes a policy in an environment. The policy can be based on a model but does not have to be. The environment can be either in simulation or in the real world. |
robopianist/execution/execution_scheduler.py |
Executes multiple model policies in the real world sequentially. |
RobotCalibration in robopianist/execution/robot_apis.py |
Can be used to test the robotics and to align the piano. |
robopianist/execution/auto_tuner.py |
Adjusts the simulation parameters to match real world observations as closely as possible. |
optuna_train.py |
Uses Optuna to optimize the divergence of training in simulation. |
All files use the parameters specified in configs.py.
In our work, we use many different libraries and frameworks. The licenses of these libraries can be found in
the licenses folder.
Our work is licensed under the MIT License. For more details, see the LICENSE file at the root of the repository.