agent_studio.agent.human_agent

Module Contents

Classes

HumanAgent

Human agents for Human-recorder

Attributes

config
logger

agent_studio.agent.human_agent.config

agent_studio.agent.human_agent.logger

class agent_studio.agent.human_agent.HumanAgent

Bases: agent_studio.agent.base_agent.BaseAgent

Human agents for Human-recorder

name = 'human'

reset(instruction: str) → None

step_action(confirmed: bool, **kwargs) → tuple[dict, bool]

Executes the code and record the result.

Parameters:

• confirmed (bool) – Whether the action is confirmed by the human.

• obs (np.ndarray | None) – The observation of the environment. For example, the screenshot.

• code (str) – The code to execute.

• annotation (dict) – The annotation of the action. For bounding box, etc.

Returns:

The result of the execution and whether the task is done.

Return type:

tuple[dict, bool]

abstract generate_action(obs: numpy.ndarray | None) → tuple[str, str]

This function shouldn’t be called by human agents.

abstract trajectory2intermediate_msg() → list[dict[str, Any]]

This function shouldn’t be called by human agents.

abstract eval(final_obs: numpy.ndarray | None = None) → dict[str, Any]

This function shouldn’t be called by human agents.