Policy Tuning and Deployment
Policy Tuning and Deployment
rlbotics turns a training run into a shippable policy - read the diagnostics, tune what matters, and export one self-contained ONNX file.
TIP
Install with pip install telekinesis-rlbotics — see Install Support for Reinforcement Learning for the simulator backends.
Tuning a Run
A reward curve tells you a run improved; the KL divergence and the action standard deviation tell you whether the next thousand iterations will improve or collapse.
Deploying a Policy
The export carries the observation normalization and, where the environment has action bounds, the action scaling - so a robot needs nothing but the file.
Where to Go Next?
Continue to the next tutorial.
Custom Environments and Algorithms
Train against your own simulator or robot, and plug in an algorithm other than PPO.
Next tutorial →