This section reports the performance of the proposed planner in handling time-critical scenarios. Two scenarios are chosen for the experiments. One requires the vehicle to perform emergency evasive manoeuvres where making a panic stop cannot avoid the collision. In the other scenario, the vehicle needs to merge into traffic, which requires careful timing and speed control.
EMERGENCY EVASIVE MANOEUVRE
In the traffic scenario shown in Figure7.10(a) and Figure 7.10(b), the vehicle is accel-erating at the hardest possible acceleration, i.e., 2m/s2, as the configuration of the cost functions gives the highest priority to the criterion of time efficiency. At the moment when the speed of the vehicle reaches 15m/s, a pedestrian appears suddenly in the center of the road in front of the vehicle. The distance between the vehicle and the pedestrian is 25m. In real life, such scenario can happen because some other obstacles blocked part of the view of the vehicle so that it could not see the pedestrian when he was still close to the roadside. According to the equations=v2/(2α), a distance of 28m is required for the vehicle to come to a stop from running at 15m/s when the hardest possible deceleration is −4m/s2. As the actual distance is 25m, it is not enough for the vehicle to implement a panic stop in order to avoid a collision with the pedestrian.
Actually, the distance between the pedestrian and the vehicle from the perspective of the planner is even smaller than 25m as the planner needs to simulate forward the vehicle dynamics system for a specific duration to compensate for the planning latency. This phenomenon can be observed by comparing the distances of the pedestrian (the cyan box) relative to the vehicle (in Figure7.10(b)) and to the green point (in Figure7.10(c)).
Consequently, the planner generates an evasive manoeuvre composed of a double lane change as demonstrated in Figure 7.10(c), Figure 7.10(d)and Figure 7.10(e). The first lane change is for evading the obstacle, while the second one is for returning back to the original travelling lane. Figure 7.11 displays the trajectories of curvature, speed, acceleration and jerk of the double lane change. In this way, the vehicle avoids potential collisions with the pedestrian. It should be pointed out that, in this experiment, the pedestrian is assumed to stand still in the road after his appearance.
(a) A pedestrian leaps into the view of the vehicle suddenly (a snapshot from the perspective of the vehicle).
(b) The trajectories that are currently executed by the vehicle at the moment when the pedestrian is located.
(c) The trajectory generated for the vehicle to swerve to avoid the pedestrian.
(d) Another trajectory generated when the vehicle executes the evasive manoeuvre.
(e) The vehicle at the end of the evasive manoeuvre.
Figure 7.10: The trajectories generated by the proposed planner in handling an emergency scenario. In each picture, the red line refers to the plan generated by the last planning cycle and is currently executed by the vehicle. The cyan line is the plan that is a planning cycle older than the red one. The green point indicates the predicted starting position of the vehicle for the red plan. The cyan box is the bounding box of the pedestrian used in the construction of the cost map of obstacles. The several layers of points are the spatial samples of the planning horizon where the red plan is generated.
(a) (b)
(c) (d)
Figure 7.11: The trajectories of curvature, speed, acceleration and jerk for the double lane change. The green and blue curves correspond to the red and cyan plans in Figure7.10(c), respectively. The short red curves record the five-second tracking result of the vehicle.
MERGING INTO MOVING TRAFFIC
Merging is among the most frequently executed complex driving manoeuvres. In the experiments shown in Figure7.12, three vehicles are running in a queue along the lane that is next to the acceleration lane. Their speeds are 30m/s, and successive vehicles keep a three-second interval. The ego-vehicle runs at 10m/s when it is about to accelerate on the acceleration lane in order to merge into the traffic on the lane to its right. If no lane is designated to the planner onto which the ego-vehicle should travel, the trajectory that is displayed in Figure7.13 is generated by the planner. Since the criterion of time efficiency is given top priority, the ego-vehicle prefers not to stay in the traffic into which it merges; it drives onto the next neighbouring lane so that it can run at a speed higher than that of the other vehicles. When the lane on which the other vehicles are running is assigned to the planner through giving a huge bonus to the candidate trajectories ending at the said lane, the vehicle ends up performing what is shown in Figure 7.14.
Figure 7.12: An example scenario where the ego-vehicle is required to merge into the traffic on the neighbouring lane.
Figure 7.15 displays the trajectories of curvature, speed, acceleration and jerk of the merging manoeuvre demonstrated in Figure 7.14(a). It can be concluded from these experiments:
• The proposed planner can generate reasonable merging behaviours under proper guidance. However, the plans generated in successive planning horizons might not be consistent as can be seen in Figure 7.14(b). There are potentially several reasons for that (cf. Subsection 7.1.3). The cause of the planning inconsistency demonstrated in Figure7.14(b) is that the new plan (red) cannot be constructed in the previous planning horizon due to the limited connectivity pattern.
• A scenario reasoning module is necessary to regulate the behaviour of the planner.
• In real-life traffic, the vehicle behind the ego-vehicle will adjust its speed due to the merging manoeuvre of the ego-vehicle. Future work regarding simulation should take into account potential interactions between the traffic participants.