garage
garage copied to clipboard
A toolkit for reproducible reinforcement learning research.
Part of #1110 I think this really should use `TimeStepBatch`, but as far as I can tell, the `trainer` doesn't have a way of retrieving one of those. It only...
This is a good implementation of conditional behavior, but it makes more sense for it to live with the rest of the codebase.
Note this builds on top of the dueling and double PRs so the diff will be redundant. Also will add tests shortly.
Add PER
Closes #1489.
For the code in examples/torch/mtsac_metaworld_mt1_pick_place.py, the policy is not able to learn a good policy. After 10e6 environment steps, the success rate is still 0 and the average return is...
We're beginning to see that some algorithms need a `TrajectoryBatch` like datatype, but with additional fields. The natural way to address this in Python is probably to use subclassing. We...
e.g. parameters such as steps_per_epoch, epoch_cycles, etc, and standardize across all agorithms in the codebase A use mode that we have in garage is the ability to control how frequently...