1. One terminal for pythonAPI

make PythonAPI

1. Another terminal for Carla server

make launch

1. It will pop up an CarlaUE4 Engine Windows, then press “Launch” and “Play”.
2. autopilot 不能用 while 收data client 會crash
3. 用 buffer 收一樣會經過幾次attack 後crash
4. SteerRightmax {0000014A#F3A8000000F3A801] 最後一個bit應為0, 用來label 為attack msg 使用（training 記得改回來）

CAN IDS Training Result Record

1. Real traffic not only ML f1-score
   1. method 1: save traffic log as an input file
   2. method 2: directly dump the traffic

model.predict(np.array(row.loc[$which row][element with feature].reshape(1,-1)

1. Carla ECU ID
   1. Throttle → 0x1F5 → 501
   2. Steering → 0x14A → 330
   3. Brake → 0x1E7→ 487
   4. Gear → 0x191 → 401
   5. Speed → 0x309 → 777
2. 要做Binary Classification or Multinomial Classification ?
   1. Jehoshua → Binary
   2. Multinomial → DoS, Spoofing, Replay
   3. 嘗試Spoofing 變成Binary dataset → GOOD
3. Replay attack 如何Implement
4. Using Models
   1. Decision Tree
   2. Logistic Regression
   3. XGB
   4. SVM
   5. KNN
   6. DNN
   7. CNN
   8. Random Forest
5. 有需要做cross validation?

1. Dataset contain 5 ECU but only inject Throttle,Steering,Brake regard as attack msg.

2. Gear, Speed regard as background noise

   # of Inject msgs

   1. Throttle → 9912(label:1)
   2. Steering → 2400(right,label:4) & 36999(left,label:2)
   3. Brake → 30000(label:3)

1. Create a virtual can

   sudo modprobe vcan
   sudo ip link add dev vcan0 type vcan
   sudo ip link set up vcan0

2. can-utils

   • Install can-utils on Linux

   $sudo apt-get install can-utils

   • Dump msg on vcan0

   $candump vcan0

   • Dump msg on vcan0 to log file

   $candump vcan0 -l

   • Dump msg on vcan0 and decode it

   $candump vcan0 | cantools decode "dbc file"

   • Replay the log file

   $canplayer -I "file.log"

   • Send CA Message

   $cansend can0 123#1122334455667788

   • See the Bus state

   $ip -details -statistics link show vcan0

3. cantools