config_example.json

{

///////////////////////////////////
// Joystick control section

//system path to the file which acts as interface to joystick
"js_path": "/dev/input/js0",

//which joystick axis to use for steering
"js_axis_steer" : 0, //PS3 left analog left and right is axis 0

//which joystick axis to use for throttle
"js_axis_throttle" : 3, //PS3 right analog up and down is axis 3

//which button to toggle self driving/prediction mode
"js_button_toggle_sd" : 12, //12 is the Triangle button on PS3 SixAxis controller

//which button to toggle logging mode
"js_button_toggle_logging" : 14, //14 is the X button on PS3 SixAxis controller

//dpad up and down control the scale on the throttle for driving when in sd mode
"js_button_dpad_up" : 4,
"js_button_dpad_down" : 6,

//we ge inputput from +- js_axis_scale and then normalize them for input to robot
"js_axis_scale" : 32767.0,


///////////////////////////////////
// Camera Type Selection
// we support three camera types
// Raspicam, the CSI camera on a ribbon made for the pi
// V4L, the v4l camera device works with many webcameras
// PointGrey, any of the PointGrey/Flir cameras compatible with flycapture2

"camera_type" : "V4L", // "Raspicam", "V4L", "PointGrey"

//When mounting the camera upside down, it's helpful to flip both of these.
//Only the Raspicam supports this operation.
"CAMERA_FLIP_VERT" : 0,
"CAMERA_FLIP_HORZ" : 0,

// Capture dimensions
// Frequently the camera is constrained to certain capture resolutions.
// In most cases, code will need to change to support other resulutions.
// The post image process may need adjustment, as well as initialization.

"raspi_camera_cap_width" : 160,
"raspi_camera_cap_height" : 120,

"v4l_camera_cap_width" : 320,
"v4l_camera_cap_height" : 240,

"pg_camera_cap_width" : 640,
"pg_camera_cap_height" : 480,

// For v4l camera, use the command line tool v4l2-ctl
//  v4l2-ctl --list-formats-ext
// to help modify the code src/contib/v4l_helper/capture_raw_frames.c
// If not YUYV, you will need to also modify src/main.cpp process_image

//final image dimensions - logged and sent to prediction
"col": 160,     //width
"row" : 120,    //height
"ch" : 3,       //depth


//video for linux uses a filename for the device access
"v4l_device_name" : "/dev/video0",

//some devices can take values of 30, 37, 50, 60, 100, 125, 137, 150, 187 ( PS3 Eye camera)
"v4l_fps": 60,

//////////////////////////////////////////////////
/// Source lib HACK!!!
// for some reason the raspicam setFrameRate isnt' working. I changed
// the framerate in the library.
// raspicam/src/private/private_impl.cpp
// change: State.framerate            = 10;
// to:     State.framerate            = 60;
//
// make this change and recompile raspicam for a better framerate.
//////////////////////////////////////////////////


////////////////////////////////////
// Logging

"log_dir" : "log",

//limit data recording to this hz
"logger_fps_limit" : 60,

//steering our bot takes a -1, to 1 range
//but our NN likes larger numbers to train against
//so scale our steering output by this constant
"STEERING_NN_SCALE" : 30.0,



////////////////////////////////////
// Car Setup

//echo various information to the console about the pwm board, servos, and esc boot status
"debug_show_car_detailed_status" : 1,

//These are the 0-15 integer plug in that we use for the Adafruit Servo Hat
"ada_steering_servo_channel" : 2,
"ada_esc_motor_channel" : 4,

//the filename used to access the pwm on the i2c bus
"pwm_device_file" : "/dev/i2c-0",

//what is the address on the i2c bus of the pwm board
"pwm_ic2_address" : "0x40",

//what is the frequency of pulse width modulation our servo likes
"pwm_servo_freq": 60,

//what is max number of levels for our servo
"pwm_servo_resolution": 4096,

//These are the integer ids of the axis on the joystick for input
"steering_js_axis_id" : 0,
"throttle_js_axis_id" : 3,


//These low, hi, med values were observed by using the 
//arrows.py script to see what worked. User can adjust to suit
//their bot.
"steering_low" : 310,
"steering_hi" : 530,
"steering_mid" : 400,

//These low, hi, med values were observed by using the 
//arrows.py script to see what worked. User can adjust to suit
//their bot.
"esc_init_lo" : 200, //reverse
"esc_init_hi" : 600, //full forward
"esc_init_mid" : 400, //idle

"esc_throttle_lo" : 350, //reverse
"esc_throttle_hi" : 450, //full forward
"esc_throttle_mid" : 400, //idle



//////////////////////////////////////////
// status led settings

//gpio pin we use for status led 
"status_pin" : 23,


//////////////////////////////////////////
// predict engine settings

//port for keras prediction server image inputs
"keras_predict_server_img_port": 9090,

//port for keras prediction server control inputs
"keras_predict_server_control_port": 9190,


//////////////////////////////////////////
// shark web app settings

//port for web http service to manage shark
"web_app_server_port": 8080,

//port for web image feed
"web_image_port": 9191,

//port for web lidar feed
"web_lidar_port": 9192,

//which model do we train by default. use a path relative to the shark/web dir where we are running
"web_rel_default_model": "../models/test",

//////////////////////////////////////////
// ec2 settings

//cents per hour max price
"aws_spot_price" : "0.90",

//this path is relative to the shark/web dir where we use this config
"aws_config" : "../aws_config.json",

//this pem need to be an absolute path
"pem_filename" : "/home/tkramer/tkramer_pi.pem",

//////////////////////////////////////////
// alt train server settings

"alt_train_host" : "sd-tkramer-01",
"alt_train_user" : "tkramer",

//default local user is pi on the raspberry pi
"local_user" : "tkramer",


//////////////////////////////////////////
// model selection

//You can enable one of two lines here to select a model for training.
//Check models.py for notes on their differences.
"model_selection" : "nvidia_transposed_inputs",
//"model_selection" : "nvidia_standard_inputs",



//////////////////////////////////////////
// training settings/defaults

//how many epochs to wait before seeing no improvement
"training_patience" : 10,

//number of samples per gradient update
"training_batch_size" : 3000,

//fraction of data held out for use in validation
"training_validation_split" : 0.1,

//number of epochs to train model. command line option can override.
"training_default_epochs" : 50,

//how many images to create when augmenting. Always leave at 1, even when no
//augmentation. total image count increase by this multiple to allow for the same image
//to be augmented in different ways. command line option can override.
"training_default_aug_mult" : 1,

//Of all the images, what fraction are augmented. command line option can override.
"training_default_aug_percent": 0.5,

//which model to load for prediction, and also training, as default model
"predict_default_model" : "./models/test",

//////////////////////////////////////////
// lidar settings

//switch to turn lidar processing on and off. sometimes, even when installed,
//we may want it off to save bettery when not using it.
"lidar_enabled": 1,

//device file to open for lidar input
"lidar_dev_file" : "/dev/ttyUSB0",

//should we print return values to console
"lidar_verbose_output" : 0,



//////////////////////////////////////////
// slam settings

//should we print slam position to the console
"slam_verbose_position": 0,

//should we output slam map to web. this take some small amount of resources.
"slam_output_debug_map": 0,



//////////////////////////////////////////
// pid settings

"pid_Kp" : 20.0,
"pid_Ki" : 0.001,
"pid_Kd" : 5.0,

//////////////////////////////////////////
// debug settings

//selectively disable control loops
//enable or disable just the car control loop
"enable_pwm_car_control" : 1,

//show the frames per second for each control loop
"debug_display_fps" : 1,

//when 1, start the recording when shark starts. run the recording even though no js input.
//this is good for validating camera and logging systems.
"debug_test_recording": 0,

//when 1, run the predict loop when shark starts.
"debug_test_predict": 0,

//when 1, output all joystick inputs to console.
"debug_test_js": 0,

//when 1, output all web image status to console.
"debug_test_web": 0,

/////////////////////////////////////
// When calibrating the inputs for pwm, use this these flags to help arrive at the proper
// values for your setup. When initing the ESC, electronic speed control, there is often
// a special sequence needed. For the Traxxas slash XL-5 ESC, you need to go directly to 
// highest, then lowest, then middle. So h l m on the keyboard before it will init.
// Critically, the middle value seems to vary slightly depending on the board and voltages.
// 400 worked as a middle value for the adafruit hat on a pi3. But with the jetson tx2 and
// a smaller pwm board, I needed a value of 380 for the middle. To find the middle value,
// go high, low middle, then wander slowly up then down from 400 while watching the arming
// light on the esc until it changes to solid red.

//when 1, look for keyboard input from console to do interactive testing of pwm channel.
//can be helpful when configuring hi low values for servo
"debug_test_pwm": 0,

//which pin channel to address for debug calibration ( 0-15 if 16 set of pins )
"debug_pwm_ch": 0,

//high value to use during pwm testing
"debug_pwm_hi": 600,

//mid value to use during pwm testing.
"debug_pwm_mid": 400,

//low value to use during pwm testing
"debug_pwm_lo": 200


}