An automated vehicle has to follow a curved road, where different obstacles can hide the lane markings.
The automated vehicle starts with the pose x_ego, y_ego, theata_ego and has a velocity of v_ego. The vehicle's sensor system detects lane markings. These serve as input for the lane-keeping system. Four obstacles can obscure the view of the lane, depending on their orientation theta_obs_1, theta_obs_2, theta_obs_3, theta_obs_4. 7 different function and vehicle model parameters are varied.
The scenario is illustrated below:
| Input | Unit | Min | Max | Type | Explanation |
|---|---|---|---|---|---|
| x_ego | m | -38.5 | -37.5 | continuous | initial x coordiante of the ego |
| y_ego | m | 120.5 | 121.5 | continuous | initial y coordiante of the ego |
| theta_ego | deg | 3.7 | 4.1 | continuous | initial orientation of the ego |
| v_ego | km/h | 20 | 40 | continuous | velocity of the ego |
| theta_obs_1 | deg | -120 | -45 | continuous | orientation of obstacle 1 |
| theta_obs_2 | deg | -105 | 25 | continuous | orientation of obstacle 2 |
| theta_obs_3 | deg | 90 | 110 | continuous | orientation of obstacle 3 |
| theta_obs_4 | deg | 90 | 110 | continuous | orientation of obstacle 4 |
| f_t | s | 0.2 | 0.4 | continuous | time horizon of the lane-keeping pre-controller |
| f_p | 0.15 | 0.25 | continuous | P of the lane-keeping controller | |
| f_i | 0.002 | 0.005 | continuous | I of the lane-keeping controller | |
| f_d | 0.6 | 1.2 | continuous | D of the lane-keeping controller | |
| v_t_1 | s | 0.1 | 0.2 | continuous | first time constant of the ego's lateral dynamic |
| v_t_2 | s | 0.005 | 0.015 | continuous | second time constant of the ego's lateral dynamic |
| v_delay | s | 0.0 | 0.1 | continuous | delay of the ego's reaction |
| Output | Unit | Type | Explanation |
|---|---|---|---|
| lane_lost | binary | indicator if the ego has lost one of the lanes entirely | |
| pos_lat_err_max | m | continuous | maximum error of the lateral position |
| pos_lat_err_min | m | continuous | minimum error of the lateral position |
| theta_err_max | deg | continuous | minimum error of the ego's orientation |
| theta_err_min | deg | continuous | maximum error of the ego's orientation |
| pos_lat_abs_err_max | m | continuous | maximum absolute error of the lateral position |
Both datasets contain concrete scenarios which are evenly distributed within the input space defined over the inputs. The train_validation dataset is generated based on the Sobol sequence, the test dataset is generated based on pseudo-random numbers generated by numpy.