Skip to content

Latest commit

 

History

History
70 lines (57 loc) · 2.66 KB

README.md

File metadata and controls

70 lines (57 loc) · 2.66 KB
Raw

Hyperparamters Optimization in game of drones

The approach is inspired by the winner (the report is available on the official website). It is the improve version of genetic algorithm (though)

Prerequisite

  1. Install game of drones binaries following the instructions from the official website

airsim_neurips2019

  1. Install tensorflow object detection API (used for gate detection)

tensorflow 1 object detection API

  1. python >= 3.6

Run

Open two terminals

  1. for running airsim binaries
cd /path/to/AirSim_Qualification
./AirSimExe.sh -windowed -opengl4
  1. for running hyperparameter optimization
cd /path/to/hyperopt_ea_game_of_drones/baselines
python baseline_racer_baseline_GA.py

Result

https://docs.google.com/presentation/d/15Ji3PlZ-SBxpDjpm1BdEuDm6ERNtocofC-S8EgZiPAM/edit?usp=sharing

Example of hyperparameters found by each algorithm after 200 iterations

Map: ZhangJiajie_Medium
Number of gates: 14

Random search (remains the same after 400 iterations) @ 46.21 seconds

v: [14.69, 34.88, 26.15, 16.1, 21.91, 25.59, 29.13, 22.2, 33.3, 25.78, 13.66, 18.41, 16.63, 22.34]
a: [54.54, 34.82, 59.93, 30.88, 140.68, 146.64, 28.58, 23.0, 79.18, 53.38, 137.22, 116.31, 86.32, 103.68]
d: [4.3, 3.95, 4.31, 5.6, 4.43, 3.9, 6.2, 4.28, 4.66, 3.55, 5.36, 6.39, 5.49, 5.05]

EA @ 45.09 seconds

v: [12.0, 25.81, 24.49, 12.0, 29.9, 12.0, 32.04, 31.94, 12.0, 12.0, 12.0, 12.0, 31.17, 12.0]
a: [125.93, 110.36, 154.52, 61.49, 50.0, 124.79, 72.61, 93.77, 121.4, 117.18, 50.0, 102.14, 91.32, 50.0]
d: [3.5, 3.5, 5.08, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 2.0]

GD @ 41.60 seconds

v: [34.25, 30.08, 24.74, 19.58, 16.93, 26.46, 34.29, 24.49, 23.31, 13.39, 27.8, 12.0, 12.0, 12.0]
a: [156.32, 85.44, 130.84, 157.3, 109.55, 107.35, 107.77, 50.0, 116.54, 59.15, 77.03, 129.87, 55.24, 60.44]
d: [3.5, 3.5, 4.8, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 2.0]

PO @ 44.00 seconds

v: [18.69, 15.76, 24.38, 22.32, 31.52, 26.01, 34.3, 11.54, 20.36, 32.44, 12.0, 12.0, 12.0, 11.64]
a: [120.07, 101.66, 110.75, 152.02, 146.16, 154.36, 99.39, 133.41, 102.7, 99.54, 146.46, 50.0, 50.0, 133.33]
d: [3.5, 3.5, 5.64, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 2.0]

BO @ 40.04 seconds

v: [22.59, 32.36, 18.66, 16.79, 16.94, 27.48, 34.08, 17.88, 28.13, 14.13, 15.66, 32.25, 26.14, 23.5]
a: [151.75, 57.78, 91.54, 131.0, 58.94, 152.9, 132.63, 113.68, 91.82, 97.83, 137.93, 68.08, 153.08, 42.96]
d: [4.21, 5.61, 4.4, 5.85, 4.42, 5.25, 3.69, 6.3, 3.9, 5.64, 4.48, 4.69, 4.42, 2.0]