| """RSL-RL PPO trainer stub (Isaac Lab ``OnPolicyRunner``). |
| |
| Hyperparameters come from the Gym task's ``rsl_rl_cfg_entry_point``. |
| Swap this class when we lock the algorithm (plain PPO vs joint residual, etc.). |
| """ |
|
|
| from __future__ import annotations |
|
|
| from typing import Any |
|
|
| from .base import Algorithm |
|
|
|
|
| class RslRlPpoTrainer(Algorithm): |
| def __init__(self, agent_cfg: Any): |
| self.agent_cfg = agent_cfg |
| self._runner = None |
|
|
| def train(self, env: Any, log_dir: str) -> None: |
| from isaaclab.utils import class_to_dict |
| from rsl_rl.runners import OnPolicyRunner |
|
|
| self._runner = OnPolicyRunner( |
| env, class_to_dict(self.agent_cfg), log_dir=log_dir, device=str(env.unwrapped.device) |
| ) |
| self._runner.learn( |
| num_learning_iterations=self.agent_cfg.max_iterations, |
| init_at_random_ep_len=True, |
| ) |
|
|
| def load(self, checkpoint: str) -> None: |
| if self._runner is None: |
| raise RuntimeError("call train() or construct a runner before load()") |
| self._runner.load(checkpoint) |
|
|
| def act(self, obs: Any) -> Any: |
| if self._runner is None: |
| raise RuntimeError("no runner") |
| return self._runner.alg.act(obs) |
|
|
