def extract_nonplayer_transform(npmes, include_people=False):
    if npmes.HasField('vehicle'):
        return transform.Transform(npmes.vehicle.transform)
    elif npmes.HasField('pedestrian') and include_people:
        return transform.Transform(npmes.pedestrian.transform)
    else:
        return None
def get_rectifying_player_transform(player_measurements):
    """ Build transform to correct for pitch or roll of car. """
    ptx = player_measurements.transform
    p_rectify = transform.Transform()
    pT = Translation(0, 0, 0)
    # Undo the pitch and roll (not yaw!)
    pR = Rotation(-ptx.rotation.pitch, 0, -ptx.rotation.roll)
    # pR = Rotation(ptx.rotation.pitch, 0, ptx.rotation.roll)
    p_rectify.set(pT, pR)
    return p_rectify
Beispiel #3
0
def run_episode(client,
                streaming_loader,
                model,
                phi,
                future,
                midlow_controller,
                car_pid_controllers,
                plottable_manager,
                waypointerconf,
                episode_params,
                metrics,
                fig,
                axes,
                cfg):
    dimconf = cfg.dim
    mainconf = cfg.main
    dataconf = cfg.data
    plotconf = cfg.plotting
    trackersconf = cfg.trackers
    expconf = cfg.experiment
    waypointerconf = cfg.waypointer
    del cfg
    
    # Build dirs.
    directory, plot_dir, dim_feeds_dir = set_up_directories(episode_params, cfg=plotconf, mainconf=mainconf, dataconf=dataconf)

    # Dump some objects.
    save_metadata(episode_params=episode_params, directory=directory, metrics=metrics)

    # Print the metrics.
    if episode_params.episode > 0: log.info("Current overall metrics: {}".format(metrics))

    # Randomize the environment.
    episode_params.settings.randomize_seeds()
    
    if midlow_controller is not None: midlow_controller.reset()
    if model is not None:
        assert(phi is model.phi)

    # Now we load these settings into the server. The server replies
    # with a scene description containing the available start spots for
    # the player. Here we can provide a CarlaSettings object or a
    # CarlaSettings.ini file as string.
    log.info("Loading the settings into the server...")
    scene = client.load_settings(episode_params.settings)
    
    # Hang out?
    time.sleep(0.01)

    # Notify the server that we want to start the episode at the
    # player_start index. This function blocks until the server is ready
    # to start the episode.
    client.start_episode(choose_player_start(scene))

    have_specific_episodes = len(expconf.specific_episodes) > 0
    if not have_specific_episodes:
        # Possibly skip the episode.
        if episode_params.episode < expconf.skip_to:
            log.warning("SKIPPING EPISODE {}\n".format(episode_params.episode) * 10)
            return
    else:
        log.info("Filtering episodes to '{}'".format(expconf.specific_episodes))
        if episode_params.episode not in expconf.specific_episodes:
            log.info("{} not in specific episodes".format(episode_params.episode))
            return
    
    # Reset the plotting.
    if plotconf.plot: _ = [ax.cla() for ax in axes.ravel()]
    plot_state = cplot.PlotState.from_plottables(plotconf, plottable_manager.base_plottables) if plotconf.plot else None

    # Hang out for a bit, see https://github.com/carla-simulator/carla/issues/263
    time.sleep(4)

    # Collect the recent measurements and controls.
    measurement_buffer = collections.deque(maxlen=dataconf.measurement_buffer_length)

    # Instatiate waypointer.
    waypointer = Waypointer(waypointerconf=waypointerconf, planner_or_map_name=expconf.scene)

    # For now, just track whether the robot is stationary.
    stop_tracker = agent_util.StationaryTracker(
        size=trackersconf.stationary_window_size, thresh=trackersconf.stationary_threshold, track_others=False)
    # Use this to detect if agent is stationary longer than it should be (e.g. 100frames->10 seconds).
    # Should be beyond the max light time and before the final timeout time (stop tracker).
    stuck_tracker = agent_util.NonredStuckTracker(
        waypointer=waypointer, size=trackersconf.stuck_window_size, thresh=trackersconf.stationary_threshold)

    # If in the past N frames, the car's pose has changed by more than M degrees, then it's executing a turn.
    turn_tracker = agent_util.TurnTracker(size=trackersconf.turn_tracker_frames, thresh=trackersconf.turn_tracker_thresh)

    # Reset the controllers.
    car_pid_controllers.reset()

    # Initially not collided.
    sum_collision_impulse = 0

    # ---------------------------------------
    # TODO incorporate pasts of other agents!
    # ---------------------------------------    
    log.warning("Defaulting to a hardcoded A past of 1")    
    A_past = 5

    # Ensure all previous episodes were properly concluded.
    for em in metrics.all_episode_metrics: assert(em.concluded)
    log.info("Current number of successes and failures: {}, {}".format(
        metrics.n_successes, metrics.n_failures))

    # Create new episode metric
    metrics.begin_new_episode(episode_params)

    waypointerconf.have_planned = False

    # Step through the simulation.
    for frame in range(0, episode_params.frames_per_episode):
        log.debug("On frame {:06d}".format(frame))
        if frame == episode_params.frames_per_episode - 1:
            summary = "Ran out of episode frames! Treating as failure"
            log.warning(summary)
            metrics.conclude_episode(success=False, summary=summary)
            with open(episode_params.root_dir + "/metrics.dill".format(episode_params.episode), 'wb') as f:
                dill.dump(metrics, f)
            break
        have_control = frame > expconf.min_takeover_frames
        
        # -----------------------
        # Read the data produced by the server this frame. Record measurements.
        # -----------------------
        measurement, sensor_data = client.read_data()
        measurement_buffer.append(measurement)
        sensor_data = attrdict.AttrDict(sensor_data)
        # Prevent sensor data mutation.
        agent_util.lock_observations(sensor_data)
        transform_now  = transform.Transform(measurement.player_measurements.transform).inverse()

        if mainconf.pilot != 'auto':
            waypoints, waypoints_local = waypointer.get_waypoints_from_transform_and_measurement(transform_now, measurement)

            if waypoints is None or waypoints_local is None:
                summary = "Waypointer error! Episode failed"
                log.error(summary)
                metrics.conclude_episode(success=False, summary=summary)
                break
        else:
            waypoints, waypoints_local = None, None

        # Some measurement-only computations
        if have_control:
            # Index the measurement to check if we've stopped.
            stop_tracker.index_new_measurement(measurement)
            stuck_tracker.index_new_measurement(measurement)
            stationary, stuck = stuck_tracker.is_stuck_far_from_red_lights(ignore_reds=waypointerconf.stuck_ignore_reds)
            if stuck: log.warning("Determined that the vehicle is stuck not near a red light!")
        else:
            stationary, stuck = False, False
                
        turn_tracker.index_new_measurement(measurement)            

        # -----------------------
        # Instantiate the object that represents all observations at the current time.
        # -----------------------
        current_obs = preproc.PlayerObservations(
            measurement_buffer, t_index=-1, radius=200, A=A_past, waypointer=waypointer, frame=frame)
        # Store some tracking data in the observation.
        current_obs.is_turning = turn_tracker.is_turning()
        current_obs.is_stuck = stuck
        current_obs.is_stationary = stationary
        transform_now = current_obs.inv_tform_t  # world2local
        if frame == 0:
            start_obs = current_obs
        log.debug("Episode={}, Current frame={}".format(episode_params.episode, current_obs.frame))

        # Populate extra metrics.
        extra_metrics = {}
        traffic_light_state, traffic_light_data = waypointer.get_upcoming_traffic_light(measurement, sensor_data)

        log.debug("Upcoming traffic light is: '{}'.".format(traffic_light_state))
        current_obs.traffic_light_state = traffic_light_state
        current_obs.traffic_light_data = traffic_light_data
        extra_metrics['red_light_violations'] = waypointer.red_light_violations
        extra_metrics['intersection_count'] = waypointer.intersection_count

        if trackersconf.reset_trackers_on_red and traffic_light_state == 'RED':
            log.debug("Resetting trackers because we observed a RED light")
            stop_tracker.reset()
            stuck_tracker.reset()
            
        # Prune near and far waypoints, possibly perturb.
        if mainconf.pilot != 'auto':
            # pdb.set_trace()
            waypoints_control, waypoint_metadata = waypointer.prepare_waypoints_for_control(waypoints_local, current_obs, dimconf=dimconf)
            npu.lock_nd(waypoints_control)
        else:
            waypoints_control = None
            waypoint_metadata = {}
            
        # Reset plotter.
        plottable_manager.reset()
            
        if mainconf.pilot != 'auto':
            # Check if we have enough waypoints to continue, and if we've stopped.
            episode_state, summary = agent_util.decide_episode_state(
                waypointer, waypoints, stop_tracker, start_obs, current_obs, goal_distance_threshold=expconf.goal_distance_threshold)
            if episode_state == agent_util.EpisodeState.SUCCESS:
                metrics.conclude_episode(success=True, summary=summary)
                break
            elif episode_state == agent_util.EpisodeState.FAILURE:
                metrics.conclude_episode(success=False, summary=summary)
                break
            elif episode_state == agent_util.EpisodeState.INPROGRESS:
                pass
            else:
                raise ValueError('unknown episode state')
                
        # ------------------
        # Update the metrics
        # ------------------
        metrics.update(measurement, extra_metrics)
        if have_control:
            metrics.update_passenger_comfort(current_obs)

        # Periodically print 
        if episode_params.episode > 0 and frame % 1000 == 0:
            log.info("Intermediate print... current overall metrics: {}".format(metrics))
            
        # Check for collision.
        if check_for_collision(measurement, sum_collision_impulse, episode_params, metrics, frame, player_transform=current_obs.inv_tform_t,
                               allow_vehicles_to_hit_us_from_behind=expconf.allow_vehicles_to_hit_us_from_behind):
            # Create control anyway in case server is looking for it?
            control = autopilot_controller.noisy_autopilot(
                measurement, replan_index=dimconf.replan_period, replan_period=dimconf.replan_period, cfg=dataconf)
            # Send the control.
            client.send_control(control)
            # Quit the episode.
            break
        
        # Update current total collision impulse.
        sum_collision_impulse = measurement.player_measurements.collision_other

        # Get the feed_dict for this frame to build features / plottables.
        # TODO should be T_past, not 3.
        if frame > 3:
            fd = streaming_loader.populate_phi_feeds(
                phi=phi,
                sensor_data=sensor_data,
                measurement_buffer=measurement_buffer,
                with_bev=True,
                with_lights=True,
                observations=current_obs,
                frame=frame)
            if have_control and dataconf.save_data:
                fd_previous = streaming_loader.populate_expert_feeds(current_obs, future, frame)
                if frame % dataconf.save_period_frames == 0:
                    fn = "{}/feed_{:08d}.json".format(dim_feeds_dir, frame)
                    log.debug("Saving feed to '{}'".format(fn))
                    preproc.dict_to_json(fd_previous, fn)
                streaming_loader.prune_old(frame)
                    
        # If we have a non-autopilot controller, and enough frames                 
        if have_control and mainconf.pilot not in ('auto', 'user', 'zero'):
            prune_nearby = current_obs.is_turning and waypointerconf.drop_near_on_turn
            if stuck or prune_nearby:
                log.warning("Vehicle is stuck or turning, preparing to adjust goal likelihood")                
                waypoints_control = waypointer.get_unsticking_waypoints(waypoints_control, midlow_controller, current_obs)
            if mainconf.save_dim_feeds:
                fn = "{}/feed_{:08d}.json".format(dim_feeds_dir, frame)
                preproc.dict_to_json(fd, fn)
            if mainconf.override_pilot:
                log.debug("Building autopilot control")
                # If not in controlled mode, or not in controlled mode yet, use autopilot.     
                control = autopilot_controller.noisy_autopilot(
                    measurement, replan_index=dimconf.replan_period, replan_period=dimconf.replan_period, cfg=dataconf)
                approach_plotting_data = None
                have_control = False
            else:
                # Generate the control for the pilot.
                control, approach_plotting_data = generate_control(
                    mainconf.pilot,
                    midlow_controller=midlow_controller,
                    car_pid_controllers=car_pid_controllers,
                    model=model,
                    fd=fd,
                    measurement=measurement,
                    waypoints_local=waypoints_control,
                    transform_now=transform_now,
                    waypointer=waypointer,
                    current_obs=current_obs,
                    sensor_data=sensor_data,
                    waypoint_metadata=waypoint_metadata)
        else:
            log.debug("Building autopilot control")
            # If not in controlled mode, or not in controlled mode yet, use autopilot.     
            control = autopilot_controller.noisy_autopilot(
                measurement, replan_index=dimconf.replan_period, replan_period=dimconf.replan_period, cfg=dataconf)
            approach_plotting_data = None

        log.debug("Control: {}".format(control))
        client.send_control(control)
        waypointerconf.have_planned = have_control

        if plotconf.plot and frame > 3:
            plot_data = cplot.GenericPerFramePlottingData(
                pilot=mainconf.pilot,
                hires=plotconf.hires_plot,
                measurement=measurement,
                feed_dict=fd,
                waypoints_local=waypoints_local,
                waypoints_control=waypoints_control,
                waypoint_metadata=waypoint_metadata,
                transform_now=transform_now,
                current_obs=current_obs,
                control=control)

            if have_control:
                cplot.update_pilots_plot(mainconf.pilot, plottable_manager, plot_data, approach_plotting_data)
            else:
                cplot.update_pilots_plot('auto', plottable_manager, plot_data, approach_plotting_data)
            
            # plottable_manager.update_from_observation(current_obs, plot_data=plot_data, control=control)
            log.debug("Plotting data")
            cplot.online_plot(model=model,
                              measurements=measurement,
                              sensor_data=sensor_data,
                              overhead_lidar=fd[phi.overhead_features][0,...,0],
                              overhead_semantic=None,
                              plottables=plottable_manager.plottables,
                              fd=fd,
                              axes=axes,
                              fig=fig,
                              plot_state=plot_state)
            if plotconf.save_plots:
                log.debug("Saving plot")
                # pdb.set_trace()
                # fig.savefig('{}/plot_{:08d}.jpg'.format(plot_dir, frame), dpi=180)

                # Get the extent of the specific axes we'll save.
                if plotconf.remove_second_row:
                    joint_extent = cplot.full_extent([axes[1,0], axes[0,1]])
                    extent0 = joint_extent.transformed(fig.dpi_scale_trans.inverted())
                    # fig.savefig('{}/plot_{:08d}.png'.format(plot_dir, frame), bbox_inches=extent0, dpi=cfg.dpi*2)
                    fig.savefig('{}/plot_{:08d}.{}'.format(plot_dir, frame, plotconf.imgfmt), bbox_inches=extent0, dpi=plotconf.dpi*2)
                else:
                    # fig.savefig('{}/plot_{:08d}.png'.format(plot_dir, frame), dpi=plotconf.dpi)
                    fig.savefig('{}/plot_{:08d}.{}'.format(plot_dir, frame, plotconf.imgfmt), dpi=plotconf.dpi)
        else:
            pass

        if frame % episode_params.snapshot_frequency == 0:
            log.info("Saving visual snapshot of environment. Frame={}".format(frame))
            # Serialize the sensor data to disk.
            for name, datum in sensor_data.items():
                if name == 'CameraRGB':
                    filename = os.path.join(directory, '{}_{:06d}'.format(name, frame))
                    datum.save_to_disk(filename)
    def __init__(self,
                 measurements,
                 t_index,
                 radius,
                 A=None,
                 agent_id_ordering=None,
                 empty_val=200,
                 waypointer=None,
                 frame=None):
        """This objects holds some metadata and history about observations. It's used as a pre-preprocessing object to pass around 
        outside of the model.

        :param measurements: list of measurements
        :param t_index: 
        :param radius: if multiagent, radius inside which to include other agents
        :param multiagent: whether to include other agents
        :param A: 
        :param agent_id_ordering: optional, None or dict str id : index of prediction.
        :returns: 
        :rtype: 

        """
        self.t = t_index

        # The sequence of player measurements.
        self.player_measurement_traj = [
            _.player_measurements for _ in measurements
        ]
        # The transform at 'now'
        self.tform_t = transform.Transform(
            self.player_measurement_traj[t_index].transform)
        # The inverse transform at 'now'
        self.inv_tform_t = self.tform_t.inverse()
        # The sequence of player forward speeds.
        self.player_forward_speeds = np.asarray(
            [_.forward_speed for _ in self.player_measurement_traj])

        # The sequence of player accelerations. These are in world coords.
        self.accels_world = np.asarray([
            vector3_to_np(_.acceleration) for _ in self.player_measurement_traj
        ])
        # Apply transform, undo translation.
        self.accels_local = (
            self.inv_tform_t.transform_points(self.accels_world) -
            self.inv_tform_t.matrix[:3, -1])
        # (T,3) sequence of player positions (CARLA world coords).
        self.player_positions_world = np.asarray([
            vector3_to_np(_.transform.location)
            for _ in self.player_measurement_traj
        ])
        # (T,3) sequence of player positions (CARLA local coords / R2P2 world frames of transform at 'now')
        self.player_positions_local = self.inv_tform_t.transform_points(
            self.player_positions_world)

        # global player rotations.
        self.rotations = np.asarray([
            transforms_rotation_to_np(_.transform)
            for _ in self.player_measurement_traj
        ])

        # Get current measurement
        self.measurement_now = measurements[self.t]

        if agent_id_ordering is None:
            # Get agents currently close to us.
            self.all_nearby_agent_ids = get_nearby_agent_ids(
                self.measurement_now, radius=radius)
            self.nearby_agent_ids = self.all_nearby_agent_ids[:A - 1]
            # Extract just the aid from (dist, aid) pairs.
            self.nearby_agent_ids_flat = [_[1] for _ in self.nearby_agent_ids]
            # Store our ordering of the agents.
            self.agent_id_ordering = dict(
                tuple(reversed(_))
                for _ in enumerate(self.nearby_agent_ids_flat))
        else:
            # Expand radius so we make sure to get the desired agents.
            self.all_nearby_agent_ids = get_nearby_agent_ids(
                self.measurement_now, radius=9999999 * radius)
            self.nearby_agent_ids = [None] * len(agent_id_ordering)
            for dist, aid in self.all_nearby_agent_ids:
                if aid in agent_id_ordering:
                    # Put the agent id in the provided index.
                    self.nearby_agent_ids[agent_id_ordering[aid]] = (dist, aid)

        # Extract all agent transforms. TODO configurable people.
        self.npc_tforms_unfilt = extract_nonplayer_transforms_list(
            measurements, False)

        # Collate the transforms of nearby agents.
        self.npc_tforms_nearby = collections.OrderedDict()
        self.npc_trajectories = []
        for dist, nid in self.nearby_agent_ids:
            self.npc_tforms_nearby[nid] = self.npc_tforms_unfilt[nid]
            self.npc_trajectories.append(
                [transform_to_loc(_) for _ in self.npc_tforms_unfilt[nid]])
        self.all_npc_trajectories = []
        for dist, nid in self.all_nearby_agent_ids:
            self.all_npc_trajectories.append(
                [transform_to_loc(_) for _ in self.npc_tforms_unfilt[nid]])

        history_shapes = [
            np.asarray(_).shape for _ in self.all_npc_trajectories
        ]
        different_history_shapes = len(set(history_shapes)) > 1
        if different_history_shapes:
            log.error(
                "Not all agents have the same history length! Pruning smallest agents until there's just one shape"
            )
            while len(set(history_shapes)) > 1:
                history_shapes = [
                    np.asarray(_).shape for _ in self.all_npc_trajectories
                ]
                smallest_agent_index = np.argmin(history_shapes, axis=0)[0]
                self.all_npc_trajectories.pop(smallest_agent_index)

        # The other agent positions in world frame.
        self.agent_positions_world = np.asarray(self.npc_trajectories)
        # N.B. This reshape will fail if histories are different sizes.
        self.unfilt_agent_positions_world = np.asarray(
            self.all_npc_trajectories)
        self.n_missing = max(A - 1 - self.agent_positions_world.shape[0], 0)

        # length-A list, indicating if we have each agent.
        self.agent_indicators = [
            1
        ] + [1] * len(self.npc_trajectories) + [0] * self.n_missing

        if self.n_missing == 0:
            pass
        elif self.n_missing > 0:
            # (3,)
            faraway = self.player_positions_world[-1] + 500
            faraway_tile = np.tile(faraway[None, None],
                                   (self.n_missing, len(measurements), 1))
            if self.n_missing == 0:
                pass
            elif self.n_missing == A - 1:
                self.agent_positions_world = faraway_tile
            else:
                self.agent_positions_world = np.concatenate(
                    (self.agent_positions_world, faraway_tile), axis=0)

        self.player_position_now_world = self.player_positions_world[self.t]
        oshape = self.agent_positions_world.shape
        uoshape = self.unfilt_agent_positions_world.shape
        apw_pack = np.reshape(self.agent_positions_world, (-1, 3))
        uapw_pack = np.reshape(self.unfilt_agent_positions_world, (-1, 3))

        # Store all agent current positions in agent frame.
        self.unfilt_agent_positions_local = np.reshape(
            self.inv_tform_t.transform_points(uapw_pack), uoshape)

        if A == 1:
            self.agent_positions_local = np.array([])
            self.agent_positions_now_world = np.array([])
            self.all_positions_now_world = self.player_position_now_world[None]
        else:
            self.agent_positions_local = np.reshape(
                self.inv_tform_t.transform_points(apw_pack), oshape)
            self.agent_positions_now_world = self.agent_positions_world[:,
                                                                        self.t]
            self.all_positions_now_world = np.concatenate(
                (self.player_position_now_world[None],
                 self.agent_positions_now_world),
                axis=0)
        assert (self.all_positions_now_world.shape == (A, 3))

        if self.n_missing > 0:
            log.warning(
                "Overwriting missing agent local positions with empty_val={}!".
                format(empty_val))
            self.agent_positions_local[-self.n_missing:] = empty_val

        self.yaws_world = [self.tform_t.yaw]

        # Extract the yaws for the agents at t=now.
        for tforms in self.npc_tforms_nearby.values():
            self.yaws_world.append(tforms[self.t].yaw)
        self.yaws_world.extend([0] * self.n_missing)

        assert (self.agent_positions_world.shape[0] == A - 1)
        assert (self.agent_positions_local.shape[0] == A - 1)
        assert (len(self.yaws_world) == A)
        assert (len(self.agent_indicators) == A)

        if waypointer is not None:
            # Get the light state from the most recent measurement.
            self.traffic_light_state, self.traffic_light_data = waypointer.get_upcoming_traffic_light(
                measurements[-1], sensor_data=None)
        else:
            log.warning("Not recording traffic light state in observation!")

        self.player_destination_world = waypointer.goal_position

        if self.player_destination_world is not None:
            self.player_destination_local = self.inv_tform_t.transform_points(
                self.player_destination_world[None])[0]
        else:
            self.player_destination_local = None
        self.yaws_local = [
            yaw_world - self.yaws_world[0] for yaw_world in self.yaws_world
        ]
def get_rectified_player_transform(player_measurements):
    return (get_rectifying_player_transform(player_measurements) *
            transform.Transform(player_measurements.transform))