Plan horizon, 0.4 s to 3 s; the paper uses 1.2 s. In Auto the model sets the parameter and runs the cases of the paper; in Hold the slider sets it. The line below the control says what the model does at this moment. With reduced motion the model shows a fixed frame, and the controls still work.
Lab · Live model · 05
A plan only at an event.
The vehicle follows a coastline on the plan that it computed from its last measurement, in an open loop. It computes a new plan only when the measured state departs from the predicted state by more than the bound, or when the horizon ends. Where the coastline turns, the prediction fails sooner and the plans follow each other more closely.
What the drawing shows
The view is from above. The coastline scrolls beneath the craft, with the sea and its depth contours off the shore. The craft is an octorotor, which is the vehicle the paper flew on the coastline. The rectangle around it is the frame of its downward camera on the ground, with the ratio of the paper's camera. In that frame the network detects the contour, drawn in accent, and its bounding box; the four corners of the box are the features of the paper's Fig. 5. The dashed band across the middle of the frame is the desired box, and the distance between the two boxes is the image error.
The trail behind the craft is the path it flew, and each dot on it is a plan. The dashed line ahead is the rest of the plan in memory: where the controller predicts the craft will go. Its model takes the contour ahead as straight, so where the coastline turns the dashed line leaves the shore, and that is what triggers the next plan. The chart is the image error against time, with a mark on its axis at each event, in the manner of the paper's Fig. 7.
The model
The plan is valid for a horizon. The craft follows it in an open loop, and it plans again at an event:
the event: the measured state departs from the predicted one by more than the bound
the other event: the horizon ends
the state: the offset of the craft from its standoff off the contour
12 steps of 0.1 s in the paper; H is the height of the canvas
This is the triggering condition of the paper: the error between the predicted and the real state, against a bound that scales with the tracking error. Between two events the control sequence runs as it was computed, and the optimisation is not solved. The measurement carries a small noise, as the visual tracking of the paper does. With this rule the paper solved its optimisation 47 per cent less often on the coastline.
The control
The slider holds the horizon, which is the longest time a plan stays valid, and therefore the largest space between two events. A short horizon gives many plans. A long horizon gives few, and a longer dashed line, and the prediction has more time to leave the shore before the error event brings the craft back.
Where it comes from
The scene draws the background of the paper An Event-Triggered Visual Servoing Predictive Control Strategy for the Surveillance of Contour-Based Areas using Multirotor Aerial Vehicles.