1. flash the TX2

1.1 download jetpack L4T 3.0

Follow flash-small-TX2 to flash the TX2

Warning:

  • do not install OpenCV4Tegra V2.4 for TX2, we will use OpenCV3
  • for TX2 you may need to flash it tiwce, one just flash the OS, the other install cuda
  • if the tx1/tx2 cannot boot the GUI, but shut down after finish loading BIOS, try to change a more powerful power source

After flash, the system can be launch.

username: nvidia

password: nvidia

2. Install basic tools

connect host and TX2 on the same local network, install arp-scan and run: `sudo arp-scan --interface=eth0 192.168.1.0/24 | grep NVIDIA' find your TX2 device.

ssh to your TX2 and run:

sudo apt-get install git

mkdir setup

cd setup

git clone https://github.com/gaowenliang/TX2-TX1-setup.git

sudo ./install.sh

sudo reboot

to install USB device and reboot.

run one by one:

sudo apt-get install terminator -y

sudo apt-get install cmake -y

sudo apt-get install vim -y

sudo apt-get install htop

3. Install ROS

follow ROS

follow installROSTX2

4. Eigen manuly

install Eigen stable release V3.3.3

follow eigen

  • it will fail 6 ctest cases out of 798
  • or if you build BTL, fail 13 ctest cases out of 832

host:

scp ~/Downloads/eigen-eigen-67e894c6cd8f.tar.bz2 ubuntu@<YOUR_SSH PATH>:~/

TX2: cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local -DCMAKE_INSTALL_PREFIX=/usr/local -DEIGEN_TEST_CXX11=ON -DEIGEN_CUDA_COMPUTE_ARCH=62 -DEIGEN_BUILD_BTL=ON

make check this step will take really a long time. Not necessary to check.

Then run: sudo make install

5. ceres manly

normal install, without eigen, need pass all ctest cases

typical Ceres test bench mark is 131s for TX2

ceres-solver.org/ceres-solver-1.12.0.tar.gz

sudo apt-get install libgoogle-glog-dev

sudo apt-get install libatlas-base-dev -y

sudo apt-get install libsuitesparse-dev -y

tar zxf ceres-solver-1.12.0.tar.gz

mkdir ceres-bin

cd ceres-bin

cmake ../ceres-solver-1.12.0 -DCMAKE_INSTALL_PREFIX=/usr/local

make -j4

make test

make install

6. Download buildOpencvTX2, remove libeigen3-dev, install Opencv manuly

follow buildOpenCVTX2 in opencv V3.2.0 opencv_test_cudev will fail, but acturally, it success, ctest made a wrong test compare between 2.0 and 2.

SDL

`sudo apt-get install libsdl-image1.2-dev

and

sudo apt-get install libsdl-dev

some useful scropts

monitor the CPU temperture

watch -n 1 cat /sys/devices/virtual/thermal/thermal_zone1/temp

check CUDA version

nvcc -V

use Manifold2 board USBs

test Ethernet speeed

$ sudo apt-get install iperf
#start server
$ sudo iperf -s
#start client
$ sudo iperf -c 192.168.1.xxx -i 5

install SSH keys

ssh-copy-id user@123.45.56.78

bash tools

devlist

devlist is

aliased to `sudo arp-scan --interface=eth0 192.168.1.0/24 | grep NVIDIA'

dev

devlist is aliased to `sudo arp-scan --interface=eth0 192.168.1.0/24 | grep NVIDIA' dev is a function

dev () 
{ 
    local ip="$(sudo arp-scan --interface=eth0 192.168.1.0/24 | grep NVIDIA | sed -n $1p | cut -d'	' -f 1)";
    echo "ssh into $ip";
    ssh nvidia@$ip
}

max_performance.sh

nvidia@tegra-ubuntu:~$ cat max_performance.sh 
echo "Enabling fan for safety..."
if [ ! -w /sys/kernel/debug/tegra_fan/target_pwm ] ; then
echo "Cannot set fan -- exiting..."
fi
echo 255 > /sys/kernel/debug/tegra_fan/target_pwm

nvpmodel -m 0
./jetson_clocks.sh