# Maximum Speed: applicable when test speed not much higher than training vel_max: 6.0 # the given weights perform smoothly at speeds between 0 - 6 m/s # segment_time = 2 * radio / vel_max # IMPORTANT TRAINING PARAM: weight of penalties (6m/s) ws: 0.00004 # smoothness cost (reduce ws when increase vel_max in training) wc: 0.001 # collision cost wg: 0.0002 # goal cost wl: 0.00 # trajectory length cost #ws: 0.00004 #wc: 0.001 #wl: 0.02 #wg: 0.0001 # trajectory and primitive parameters horizon_num: 5 # grids num in horizon vertical_num: 3 # grids num in vertical horizon_camera_fov: 90.0 # horizon camera fov vertical_camera_fov: 60.0 # vertical camera fov horizon_anchor_fov: 30 # horizon fov of each gird vertical_anchor_fov: 30 # vertical fov of each grid goal_length: 10 # used for standardization of goal penalties (should >= 2 * radio_range) radio_range: 4.0 # planning horizon: 2 * radio_range vel_fov: 90.0 # not use currently radio_num: 1 # 1 just ok (deprecated) vel_num: 1 # 1 just ok (deprecated) vel_prefile: 0.0 # 0 just ok (deprecated) # For data efficiency, we randomly sample multiple vel and acc for each depth image with the following the distribution # values at normalized speed (actual speed can be denormalized by multiplying v_multiple) # 单位数据倍数: v_multiple = 0.5 * v_max = radio / time # v数据的均值: v_mean = v_multiple * v_mean_unit # v数据的方差: v_var = v_multiple^2 * v_var_unit # a数据的均值: v_mean = v_multiple^2 * a_mean_unit # a数据的方差: v_var = v_multiple^4 * a_var_unit vx_mean_unit: 1.5 # vel_x: skewed distribution vy_mean_unit: 0.0 vz_mean_unit: 0.0 vx_var_unit: 0.15 vy_var_unit: 0.45 vz_var_unit: 0.1 ax_mean_unit: 0.0 ay_mean_unit: 0.0 az_mean_unit: 0.0 ax_var_unit: 0.0278 ay_var_unit: 0.05 az_var_unit: 0.0278 # collision penalties alpha: 10.0 d0: 1.2 r: 0.6 # vel penalties (deprecated) alphav: 2.0 v0: 3.5 rv: 1.5 # acc penalties (deprecated) alphaa: 2.0 a0: 3.5 ra: 1.5 # vel and acc weight (deprecated) wv: 0.0 wa: 0.0