|
| 1 | +[19.0.0] |
| 2 | + * New prameters: |
| 3 | + - server::dist_noise_rate (default value: 0.0125) |
| 4 | + - server::focus_dist_noise_rate (default value: 0.0125) |
| 5 | + - server::land_dist_noise_rate (default value: 0.00125) |
| 6 | + - server::land_focus_dist_noise_rate (default value: 0.00125) |
| 7 | + |
| 8 | + * New command: |
| 9 | + - "(dash (l POWER DIR) (r POWER DIR))" |
| 10 | + This is an extension of the dash command. All version players |
| 11 | + can use this format. If the command is accepted, players can |
| 12 | + perform acceleration and direction change simultaneously based |
| 13 | + on the bipedal dash model. |
| 14 | + |
| 15 | + - "(gaussian_see)" |
| 16 | + All version players can use this command. If the command is |
| 17 | + accepted, rcssserver sent a reply message, "(ok gaussian_see)". |
| 18 | + This command is used for a gaussian noise mode describing below. |
| 19 | + |
| 20 | + * Introduce a bipedal dash model. Players can now independently |
| 21 | + issue dash commands to the left and right legs. This means that |
| 22 | + players can now apply different accelerations to each leg. With |
| 23 | + the bipedal dash model, players can perform acceleration and |
| 24 | + direction changes simultaneously, governed by differential drive |
| 25 | + kinematics. |
| 26 | + |
| 27 | + The rotation is calculated as: |
| 28 | + |
| 29 | + rotation = (left_leg_vel.bx - right_leg_vel.bx)/(player_size*2) |
| 30 | + |
| 31 | + where bx is the x-component of the vector with the player's |
| 32 | + body direction as the x-axis. |
| 33 | + |
| 34 | + The player's result velocity is calculated as: |
| 35 | + |
| 36 | + vel = (left_leg_vel + right_leg_vel)/2 |
| 37 | + |
| 38 | + Stamina is consumed independently on the left and right legs, |
| 39 | + and the total consumption is the sum of half of each leg: |
| 40 | + |
| 41 | + stamina = stamina - (left_consumed/2 + right_consumed/2) |
| 42 | + |
| 43 | + * Introduce a new gaussian observation noise model. This model is |
| 44 | + activated by "(gaussian_see)" command and replaces the previous |
| 45 | + quantization model. In this model, In this model, the noised |
| 46 | + distance in the player's observation is determined by a |
| 47 | + Gaussian distribution: |
| 48 | + |
| 49 | + stddev = actual_dist * noise_rate + focus_dist * focus_noise_rate |
| 50 | + noised_dist = max(0.0, normal_distribution(actual_dist, stddev)) |
| 51 | + |
| 52 | + where normal_distribution(mean, stddev) is a random number |
| 53 | + generator based on a Gaussian distribution with given mean and |
| 54 | + standard deviation. actual_dist represents the actual distance |
| 55 | + between the observed object and the player, while focus_dist is |
| 56 | + the distance between the observed object and the player's focus |
| 57 | + point. noise_rate and focus_noise_rate are determined by |
| 58 | + heterogeneous parameters, with default values defined as new |
| 59 | + server parameters, uniformly set for all player types in the |
| 60 | + current version. For ball and player observation, |
| 61 | + server::dist_noise_rate and server::focus_dist_noise_rate are |
| 62 | + applied, while for flags (landmark objects), server::land_dist_noise_rate |
| 63 | + and server::focus_dist_noise_rate are applied. |
| 64 | + |
| 65 | + The velocity noise formula is similar to the previous one. |
| 66 | + The formula of dir_chg remains unchanged, while dist_chg is |
| 67 | + calculated as: |
| 68 | + |
| 69 | + dist_chg = actual_dist_chg * noised_dist / actual_dist |
| 70 | + |
| 71 | + where actual_dist_chg represents the x-component of the velocity |
| 72 | + vector with the direction from the player to the observed object |
| 73 | + as the x-axis. The resulting dist_chg value is rounded to two |
| 74 | + decimal places before being sent. |
| 75 | + |
| 76 | + * Improve the JSON game log format. The format of each data has |
| 77 | + been reviewed to make it easier to parse and the JSON rcg is |
| 78 | + now recorded as a pure JSON file. The parser library is bundled |
| 79 | + in rcssmonitor. |
| 80 | + |
1 | 81 | [18.1.3]
|
2 | 82 | * Fix an issue in the penalty shootouts referee. If both teams
|
3 |
| - score the same when finishing all extended trials, the penalty |
4 |
| - shootouts referee will not end the game and the simulator will |
5 |
| - get stuck. Thanks go to Omid Amini for providing the patch. |
| 83 | + score the same when finishing all extended trials, the penalty |
| 84 | + shootouts referee will not end the game and the simulator will |
| 85 | + get stuck. Thanks go to Omid Amini for providing the patch. |
6 | 86 |
|
7 | 87 | [18.1.2]
|
8 | 88 | * Fix a problem of v18 observation noise model. Quantized distance
|
9 | 89 | values affected by the focus point are now rounded to one decimal
|
10 |
| - place. |
| 90 | + place. |
11 | 91 |
|
12 | 92 | [18.1.1]
|
13 | 93 | * Fix a problem in which the focus point is sometimes not updated.
|
|
0 commit comments