machines | Blue Duck Valley Rd

ArduRover based nppilot log

Sun, 14 Feb 2016 00:00:00 +0000

Day 1

Controller brownout when accelerating. I can reproduce this on the bench by gunning the throttle with the wheels off the ground. The documentation suggests that the power module is designed for 3S batteries, not the 2S equivalent that my 7.2 V NiMH packs are. Powering off just the cars BEC shows the same problem.

Solution: remove the power module. Add a small LiPo with a switchmode BEC to the controllers servo rails.

Oversteering at a corner.

Day 2

Plan:

Figure out how to restart the log.
Profile the speed versus demand curve.
Increase the standard speed to 4 m/s.
Reduce the steering gain to reduce oversteering.
Brake at the end of the run
See how the throttle stick changes speed on Auto
- Should be throttle_base + throttle_nudge + pid(speed error)
- PID is PID(speed error in m/s * 100) / 100
- Range is 0 .. 100
- Speed is from ground_speed, which is from either the EKF or GPS.
Calibrate centre on the steering
- startup_ground calls trim_radios at startup which stores the steering trim (yay!)
Test the throttle range
Check if the motor controller can be calibrated: seems not, but the stick position on power on sets neutral.
See if it can beep more
See when the waypoints get cleared
- Based on the trainer switch release.
- In manual, should log ‘Erasing waypoints’
  - If roll is full right then also sets home
- In training, sets waypoint. Prints ‘Learning waypoint’
- In auto, goes to RTL (return to landing)
See if the log can be restarted on a signal
- Logs if log_bitmask & MASK_LOG_WHEN_DISARMED, but never stops the log.
The main metrics are

1650=4m/s span=1500-1900 Changes:

Turn on COMPASS_LEARN
Set GPS_NAVFILTER to Automotive

To investigate:

GCS_PID_MASK: mask of PIDs to send PID_TUNING messages for
- Defaults off
- No obvious speed version. Not obvious where the message is defined.
Can arm / disarm using the rudder - right to arm, left to disarm; while the throttle is in the deadzone

NAZE32

Sun, 12 Jul 2015 00:00:00 +0000

The NAZE32 is a STM32 based flight controller. I own a rev5 version with the baro and magnetic compass that would be a great sensor and controller board for nppilot. This page has notes on the hardware itself.

The schematic is hard to find. The rev4 schematic is mirrored here. The CPU is a STM32F103CBT6 in a 48 pin package.

Basics

The two main connectors are the motor and receiver blocks. PPM input is on RX1.

UARTs

UART1: pin 30, 31 to the CP2102
UART2: pin 12, 13 to RC_CH3, RC_CH4
UART3: pin 21, 21 mapped to SCL, SDA (so unusable)
UART1: pin 42 (TX), 43 (RX) to PWM3, PWM4 (motor block)

The other remaps aren’t brought out on the 48 pin chip.

Citizen

Sun, 02 Nov 2014 00:00:00 +0000

A balancing robot written in Go and running on ARM Linux.

Mod list

Changes to make:

Move the gyro down between the wheels. Reduces the acceleration seen due to falling.

Other robots

There’s a ton out there.

lil-bot uses the same motors (MG-6-48) but with encoders. 7xAA so 8.4 V to 10.5 V. 100 Hz control loop. MPU-6050 with 2 % complementary filter. Has a err**2 term as well.

v1

v1 can balance but poorly. The big differences were:

Better pitch measurement through a complementary filter. Gives the absolutes of the accelerometers and noise rejection of the gyros.
Integral control to respond quicker to small offsets.

Things I also changed:

PWM rate up to 10 kHz
Control loop up to 100 Hz
Gyro bias removal. The gyro had a 0.34 deg/s bias.

Things to check:

The gyro scale might be wrong and might be over reporting the angle.
The deadband on the motors is quite bad and is probably hurting the small tilt response. I might be able to kick the motors to start things turning then return to the demanded speed.
Is the gyro saturating?
I2C speed
Add voltage compensation?

v0

v0 is done but doesn’t really work. Issues:

The motors have a high deadband and are too high speed.
The pitch is based off the accelerometers only and seems very noisy.
The motors are independant, so at low drive sometimes one will start and the other stay still.

Next steps:

Measure the CPU load
Measure the pitch while standing upright and steady
Investigate complementry filters to combine the gyro and accelerometers
Investigate the IMUs advanced features to see if it already gives a good pitch
Profile the motors to see if I can lower the deadband (dither?)
Check I2C rate is 100 kHz or higher.

Notes:

The 20 ms timer fires at ~16 ms and ~24 ms. HZ is 128. Turn on TCB as a clock source, switch to low latency, switch to tickless.
IMU has:
DLPF, a low pass filter. Use to filter accelerometers?
Default accelerometer range is +-2G (as the default reading is ~16000)

CopterControl

Sat, 29 Jun 2013 00:00:00 +0000

TCHAIN_PREFIX = arm-none-eabi-

export CODE_SOURCERY = NO to skip -fpromote-loop-indicies

Command line programming

There’s an experimental command line programmer in ground/openpilotgcs/src/experimental/USB_UPLOAD_TOOL. Use http://juju.net.nz/src/pilot-hacks.git/tree/opuploadtool.mk to build it.

Custom code

The link command is

arm-none-eabi-gcc ... $defines $includes $objs
--output foo.elf -nostartfiles -Wl,-Map=foo.map,--cref,--gc-sections
-lc -lm -lgcc
-T../PiOS/STM32F10x/link_STM32103CB_CC_Rev1_memory.ld
-T../PiOS/STM32F10x/link_STM32103CB_CC_Rev1_sections.ld

The firmware image has a 100 byte header generated by:

python $openpilot/make/scripts/version-info.py
--path=$openpilot
--template=$openpilot/make/templates/firmwareinfotemplate.c
--outfile=$openpilot/build/fw_coptercontrol/fw_coptercontrol.bin.firmwareinfo.c
--image=$openpilot/build/fw_coptercontrol/fw_coptercontrol.bin
--type=0x04 --revision=0x02

According to the linker script:

BL_FLASH is 0x08000000 to +0x2f7f
BD_INFO is 0x08002f80 to +0x80
FLASH is 0x08003000 onwards to (128 - 12) k
SRAM is 0x20000000 to +20 k

Vectors are at the start of user flash (…3000), live in .isr_vectors, and are from startup_stm32f10x_MD_CC.S.

Bootloader checks the blob at pios_board_info_blob (0x08002f80):

struct pios_board_info {
uint32_t magic;
uint8_t board_type;
uint8_t board_rev;
uint8_t bl_rev;
uint8_t hw_type;
uint32_t fw_base;
uint32_t fw_size;
uint32_t desc_base;
uint32_t desc_size;
uint32_t ee_base;
uint32_t ee_size;
} __attribute__((packed));

for bdinfo->fw_base & 0x2FFE0000 == 0x20000000. Sets SP to fw_base[0] and jumps to *fw_base[1]. Ah, so checks that the SP is valid.

What does the image look like? The command line flasher just takes a binary file and programs it to 0x08003000.

Hardware

The status LED is on PA6. Crystal is 8 MHz. Blink it!.

Schematic is at http://wiki.openpilot.org/download/attachments/12386767/CopterControl%20Schematic.pdf

Ports:

USART connectors are GND / PWR / TX / RX
CONN2 is USART1 on PA9 & PA10 with an invert select on PB2
CONN3 (Flexi) is USART3 on PB10 & PB11

Solar power for my office

Thu, 29 Mar 2012 00:00:00 +0000

I’m getting a 10 square metre office built in the garden. Let’s solar power it! This page has notes.

Most bits are from Jaycar.

A 380 W pure sine wave inverter. The fan stays off under normal < 100 W draw use. No inverter noise. Not hot. Thumbs up!

Started with a single 80 W 12 V Chinese monocrystaline panel. Peaks at 5 A. Peak power at ~16 V. Comes with waterproof connectors.

Started with a PWM charge controller. Pick 20 A for future capacity. Developed a peak of around 60 W (12 V @ 5 A). Changing to a MPPT gives an extra ~30 % (16 V = +4 V @ 5 A).

20 A mains flex is fine. Not UV proof though.

Added a second panel.

Thinkpad R500 laptop with Ubuntu Precise takes 40 W @ 240 V with the screen on, 60 W peak. Charge controller said it was delivering 50 W so 10 W loss + error in the inverter. Power factor is nasty.

Single small spare car battery gives OK storage. Only really needed for bridging. Probably won’t last.

Mild cloud cover reduces output to ~1/3. Surprising.

Point the panel north, your latitude from the ground. I was surprised with how accurate this is at the start of Autum. Measured the altitude of the sun using my shadow and it matched well. Solar angle calculator says 70 degrees in summer, 46 in autumn, and 22 in winter, and winter is coming…

Going for best winter performance (31 degrees) gives 2.81 kWh/m2/day at worst, 4.54 in summer. Best summer gives 2.30 to 5.64. Best winter gives 22 % more energy at the most needed time.

(Panel is 1210 x 540 = 0.65 m2. 80 W and ~1 kW/m2 gives ~12 % efficient)

8 hours with 2 x 15 W lights, 50 W laptop, 10 W loss = 90 W * 8 hours = 0.72 kWh. 0.65 x 2 m2 of panels @ 12 % @ 2.81 hWh/m2/day = 0.44 hWh = 61 % of needed.

Azimuth of the sun changes by 90 degrees during the day.

Use a 230 V relay for automatic changeover. Draws < 1 W. Change in < 16 ms.

Parts

Catalogue	Name	Price
MP3735	12V/24V 30A MPPT Solar Charge Controller	309
ZM9300	80W Recreational Solar Package Deal	375
MP-3128	Solar Charging Controller	-43.9
ZM9097	Powertech Monocrystalline Solar Panel - 80W	359
MI5162	380 Watt 12VDC to 230VAC Pure Sine Wave Inverter	312
	12 V car battery
WB1568	5 m 20 A mains flex	24.5
	Hobby box	5
	M4 x 12	6
	M6 x 15	6
	5 m aluminium channel	60

Total is $1400. Too much.

Pictures

Further

We have gas hot water and a heat pump. Consumption is 6500 kWh/year @ 20.44 c/kWh = $1335 / year. Peak is July at 22 kWh/day. Summer is 10 kWh/day.

The June 2009 Motukiekie Island ECEA study says 2.72 kWp cost $17 k. Batteries $12 k, inverters $13 k.

$17 k / 2.72 kWp = $6.25/Wh. A 80 W panel would be $500. Difference must be framing and installation.

Solarbuzz says the average price per Wp is $2.29 US for the module. Gives $7800 for the same system.

Guess and say we need a 3 kW inverter. $0.711 US/kW ~= $3000.

Auckland and Christchurch get 2050 sunshine h/year. 3.1 kWp ideally to cover it.

The iMiEV has a 16 kWh battery and can do 160 km. 20 km/day ~= 2kWh.

nppilot

Mon, 01 Jan 0001 00:00:00 +0000

Some things to try as part of fiddling:

Using golang as the main language
Using libmaple instead of the ST libraries

Investigation stages

[[nppilot/boottime]]

Parts

CPU: Arietta G25
- 3.3 V only.
- GPIO pins are 8 mA up to 3.6 V
- 85 mA @ 5 V
- On board NCP1529 as 5 V regulator. Specified 2.7 to 5.5 V
- Watchdog
- Serial: +3V, 5V, RX, TX, WKP, GND
GPS: Sparkfun LS20031 (MC-1513)
IMU: Drotek MPU6050
- 3.3 V, I2C, drivers/iio/imu/inv_mpu6050
Receiver: FrSky TFR4
- 5 V+. PPM output. G25 I/O lines can interrupt on change.
LCD: MOD-LCD3310
- 3.3 V, SPI + control, driver
- pcd8544 based? Or more recent?
Serial: Sparkfun FTDI
- GND CTS VCC TXD RXD RTX

Unknown

Arietta power supply

AT91G25 TC unit

There’s bits of a driver under drivers/misc/atmel_tclib.c and include/linux/atmel_tc.h.

Counter mode has two capture registers that can trigger from a TIOAn line. They are:

TIOA0: PA21
TIOA1: PA22
TIOA2: PA23
TIOA3: PC2
TIOA4: PC5
TIOA5: PC12

Development environment

~~GCC 4.8 with gccgo for arm-none-eabi~~
~~OpenOCD works inside VMWare~~
~~OpenOCD plus Signalyzer Lite works~~

CopterControl basics:

CopterControl programmer works
Serial to JSC cable built
Send and receive serial over the main port

Command:

Can build and run Mini XPlus kernel

Communications:

STM32-H103 shows as a USB CDC device
Measure latency and jitter against x86
Measure latency and jitter against XPlus

Using golang:

~~Build for arm-none-eabi~~
Interfacing with C functions
Interfacing with C types
When the runtime is needed

Third party components

Qualities for third party components:

Open license
Takes patches
Healthy community?

RTOS:

NuttX (BSD)
ChibiOS (GPLv3)
TNKernel (BSD-like?)

HAL:

ChibiOS
libmaple

Drivers and flight:

CoperControl / OpenPilot
CrazyFlie
Ardupilot

machines | Blue Duck Valley Rd

ArduRover based nppilot log

Day 1

Day 2

The main metrics are

NAZE32

Basics

UARTs

Citizen

Mod list

Other robots

v1

v0

CopterControl

Command line programming

Custom code

Hardware

Solar power for my office

Parts

Pictures

Further

nppilot

Investigation stages

Parts

AT91G25 TC unit

Development environment

Third party components