RLExecutor

class RLExecutor(BaseExecutor)

Default executor for reinforcement-learning agents. Implements the standard loop documented in Optimization loop: construct EnvData / AgentData, load checkpoints, then run reset → compute_action → step → experience → learn until the job completes.

You normally do not subclass RLExecutor; pass it to Connector along with your RLAgentEnv class.

Import

from adk.executors.rl import RLExecutor, RLExecutorConfig

Connector usage

from adk.connector import Connector
from adk.executors.rl import RLExecutor, RLExecutorConfig

app = Connector(
    RLExecutor,
    rl_agent_env_class=MyAgent,
    model_handler=MyAgent,
    config=RLExecutorConfig(maximum_training_steps=50_000),
)

Constructor kwargs (via Connector)

Kwarg	Description
rl_agent_env_class	Your RLAgentEnv subclass (required).
throw_errors	Propagate errors instead of sending platform error responses.
config	RLExecutorConfig instance (see below).

RLExecutorConfig

class RLExecutorConfig(BaseModel):
    maximum_training_steps: int = 100_000
    maximum_inference_steps: int = 1_000

Field	Description
maximum_training_steps	Stop training runs after this many env steps.
maximum_inference_steps	Stop inference runs after this many env steps.

Behaviour notes

• Models are loaded from models/<genie-model>/models/ at the start of each run.
• During training, checkpoints are saved every 100 episodes.
• During inference, the run finishes with Satisfied when an episode ends with terminated=True.
• Progress is reported through update_display with OptimizationStats and AgentStats.