예제 #1
0
    def handle_gym_message(self, envelope: Envelope) -> None:
        """
        Forward a message to gym.

        :param envelope: the envelope
        :return: None
        """
        gym_message = GymSerializer().decode(envelope.message)
        performative = gym_message.get("performative")
        if GymMessage.Performative(
                performative) == GymMessage.Performative.ACT:
            action = gym_message.get("action")
            step_id = gym_message.get("step_id")
            observation, reward, done, info = self.gym_env.step(
                action)  # type: ignore
            msg = GymMessage(performative=GymMessage.Performative.PERCEPT,
                             observation=observation,
                             reward=reward,
                             done=done,
                             info=info,
                             step_id=step_id)
            msg_bytes = GymSerializer().encode(msg)
            envelope = Envelope(to=envelope.sender,
                                sender=DEFAULT_GYM,
                                protocol_id=GymMessage.protocol_id,
                                message=msg_bytes)
            self._send(envelope)
        elif GymMessage.Performative(
                performative) == GymMessage.Performative.RESET:
            self.gym_env.reset()  # type: ignore
        elif GymMessage.Performative(
                performative) == GymMessage.Performative.CLOSE:
            self.gym_env.close()  # type: ignore
예제 #2
0
파일: env.py 프로젝트: mattmcd/aea-examples
    def _decode_percept(self, envelope: Envelope,
                        expected_step_id: int) -> Message:
        """
        Receive the response from the gym environment in the form of an envelope and decode it.

        The response is a PERCEPT message containing the usual 'observation', 'reward', 'done', 'info' parameters.

        :param expected_step_id: the expected step id
        :return: a message received as a response to the action performed in apply_action.
        """
        if envelope is not None:
            if envelope.protocol_id == 'gym':
                gym_msg = GymSerializer().decode(envelope.message)
                gym_msg_performative = GymMessage.Performative(
                    gym_msg.get("performative"))
                gym_msg_step_id = gym_msg.get("step_id")
                if gym_msg_performative == GymMessage.Performative.PERCEPT and gym_msg_step_id == expected_step_id:
                    return gym_msg
                else:
                    raise ValueError(
                        "Unexpected performative or no step_id: {}".format(
                            gym_msg_performative))
            else:
                raise ValueError("Unknown protocol_id: {}".format(
                    envelope.protocol_id))
        else:
            raise ValueError("Missing envelope.")
예제 #3
0
    def close(self) -> None:
        """
        Close the environment.

        :return: None
        """
        gym_msg = GymMessage(performative=GymMessage.Performative.CLOSE)
        gym_bytes = GymSerializer().encode(gym_msg)
        envelope = Envelope(to=DEFAULT_GYM, sender=self._agent_public_key, protocol_id=GymMessage.protocol_id,
                            message=gym_bytes)
        self._agent.outbox.put(envelope)
        self._disconnect()
예제 #4
0
    def _encode_action(self, action: Action, step_id: int) -> Envelope:
        """
        Encode the 'action' sent to the step function as one or several envelopes.

        :param action: the action that is the output of an RL algorithm.
        :param step_id: the step id
        :return: an envelope
        """
        gym_msg = GymMessage(performative=GymMessage.Performative.ACT, action=action, step_id=step_id)
        gym_bytes = GymSerializer().encode(gym_msg)
        envelope = Envelope(to=DEFAULT_GYM, sender=self._agent_public_key, protocol_id=GymMessage.protocol_id,
                            message=gym_bytes)
        return envelope
예제 #5
0
    def reset(self) -> None:
        """
        Reset the environment.

        :return: None
        """
        if not self._agent.multiplexer.is_connected:
            self._connect()
        gym_msg = GymMessage(performative=GymMessage.Performative.RESET)
        gym_bytes = GymSerializer().encode(gym_msg)
        envelope = Envelope(to=DEFAULT_GYM, sender=self._agent_public_key, protocol_id=GymMessage.protocol_id,
                            message=gym_bytes)
        self._agent.outbox.put(envelope)
예제 #6
0
    def reset(self) -> None:
        """
        Reset the environment.

        :return: None
        """
        self._step_count = 0
        self._is_rl_agent_trained = False
        gym_msg = GymMessage(performative=GymMessage.Performative.RESET)
        gym_bytes = GymSerializer().encode(gym_msg)
        envelope = Envelope(to=DEFAULT_GYM,
                            sender=self._skill_context.agent_public_key,
                            protocol_id=GymMessage.protocol_id,
                            message=gym_bytes)
        self._skill_context.outbox.put(envelope)
예제 #7
0
    def handle_envelope(self, envelope: Envelope) -> None:
        """
        Handle envelopes.

        :param envelope: the envelope
        :return: None
        """
        gym_msg = GymSerializer().decode(envelope.message)
        gym_msg_performative = GymMessage.Performative(
            gym_msg.get("performative"))
        if gym_msg_performative == GymMessage.Performative.PERCEPT:
            assert self.context.tasks is not None, "Incorrect initialization."
            assert len(self.context.tasks) == 1, "Too many tasks loaded!"
            gym_task = cast(GymTask, self.context.tasks[0])
            gym_task.proxy_env_queue.put(gym_msg)
        else:
            raise ValueError(
                "Unexpected performative or no step_id: {}".format(
                    gym_msg_performative))