Closed sujithvemi closed 5 years ago
Hi @HongyiSun and @kechxu is it because of the assumption that a vehicle will accelerate once it is leaving the curve, that the prediction is only till max curvature point if the max curvature is over 0.14? Can you please confirm if my hypothesis is correct?
@sujithvemi Thanks for your asking. As to the current trajectory generation method, we need to find an end state for polynomial fit. Typically, vehicles decelerate while it is entering a turning. Thus, we select the most representative point -- the point with the maximal curvature as the end state and set the speed at the end state as a pre-defined comfortable speed.
In move_sequence_predictor, while fitting the longitudinal polynomial, we predict an end state in the longitudinal direction. This computation is done using the curvature information, to predict the velocity for the end state. There are multiple cases in this particular function, most of them return the estimated time as being equal to the prediciton horizon. In one case it returns the time that will be taken to reach the point with maximum curvature. Can someone explain the reasoning behind this? Look at the code below to understand what I am referring to.
Edit: I understand that the prediction is only done till the max_curvature point in the scenario I was looking at. But I am curious to understand why this is the case, as the trajectory predicted when the obsctacle is very close to this point is very short and is not reliable for planning.