4x4.md

#Rough concept Waterproof box, half decent motors, some extrusion

BOM (provisonal)

• $190 i7 industrial 8Gb or $420 Open(ish) mainboard
• $180 Bldc x2 inc 3200PPR internal encoder
• ~$40? Twisted fields Open hardware Driver board
• $50 18650 battery pack& Charger XT60 280x 68 x 40mm-
• $50 Waterproof lectronics case SO41 External: 345x268x120mm Internal: 311x210x105mm
• $10 36v > 12v
• $30 ESP32 Feather
• $25 base plate x2 this joins the extrusion and the Peli clone sits on it
• $12 2x 30mm
• $14 2x 40mm
• $12 caster for 2wd

Upgrades

• $199 OAK-D
• Ardusimple RTK
• Tyre
• Additional motors for 4x4 inc 3200PPR internal encoder
• Additional driver for 4x4
• Bigger wheels

Base software stack

• Linorobot2
• linorobot video
• Rosboard
• Groot via WebGL

Similar platforms

• Robaka ROS1, 2WD SerdaAbdali fork
• Morph ROS1 became Ovo (evaluating Linorobot2 for ROS2 work)
• AgileX scout mini IP22, expensive
• Rover robotics no BLDC
• LeoRover ROS1/2 no BLDC.
• Ubiquity ROS1

Ag use case

• 75cm wide deep compost mulch beds. (as common in Agroecological production)
• Weeding needed in babyleaf lettuce and brassica.
• Bed layout: <10cm wheel> <22.5 bed area> <10cm wheel> <22.5 bed area> <10cm wheel>
• 1x row of crop in centre of each 22.5cm bed area.
• Nvidia team
• Tensorflow model to follow crop rows - Outputs Cmd_vel

The core of this navigation strategy is the VisualServoing

Ag use case: weeder actuator hardware

• Linear actuator openbuilds
• IP67 rated Nema23
• Laser from Neje 20w

Ag use case: Initial actuator CV

• Todo: CV to control robot speed. When green is detected between the crop row; slow/stop, move laser to detected position, burn weed. Then continue.
• A Survey of Deep Learning Techniques for Weed Detection from Images 2021
• Artificial intelligence for weed detection
• CV datasets

Arrangement of 75cm bed, showing two rows of crops, rover & paths for rover.

Ready for GUI programming Or AI computer vision

• Service robots
• Lawn mowers
• R&D

To do

• I2C commander supported by Linorobot
• RTK
• Flexbe
• TOF sensors
• Foxglove studio
• Dual camera

RTK

• £233 Cubepilot RTK
• £300 H-RTK
• £80 PX2.4.8
• Via MavROS

Orbslam3

sudo apt update sudo apt install python3-opencv libopencv-dev libeigen3-dev ros-$ROS_DISTRO-vision-opencv && sudo apt install ros-$ROS_DISTRO-message-filters ninja-build catch2

Get Pangolin

mkdir pangolin cd pangolin git clone --recursive https://github.com/stevenlovegrove/Pangolin.git cd Pangolin

Install dependencies (as described above, or your preferred method)

./scripts/install_prerequisites.sh cmake -B build -GNinja cmake --build build

Run me some tests! (Requires Catch2 which must be manually installed on Ubuntu.)

ctest

Then git clone https://github.com/zang09/ORB-SLAM3-STEREO-FIXED.git ORB_SLAM3 cd ORB_SLAM3 chmod +x build.sh ./build.sh

cd ~/colcon_ws/src git clone https://github.com/zang09/ORB_SLAM3_ROS2.git orbslam3_ros2

Ref https://github.com/zang09/ORB_SLAM3_ROS2#how-to-build

OS

• Pi 20.04
• Jetson 20.04
• Then install ROS2me It would be nice if the ROS2ME script supported 22.04/ Humble.

Interesting packages.

• USB Cam DepthAI
• robot_Localisation info
• OrbSLAM3
• Rosboard

CM4 based compute

• Radxa CM3
• [Gumstix carrier] (https://github.com/gumstix/PKG900000001400)

RiscV

• Compute https://www.sifive.com/cores/performance-p650 https://github.com/siemens/isar-riscv/blob/main/doc/ROS2.md
• GPU https://www.think-silicon.com/neox-graphics
• MCU (Zephyr FreeRTOS) https://github.com/stnolting/neorv32 on Orangecrab https://www.tindie.com/products/greeeg/orange-crab/
• NPU 2.5 tops https://www.aliexpress.com/item/1005004332478616.html
• P920 256 TOPS https://riscv.org/blog/2021/07/neuralscale-industry-leading-general-purpose-programmable-npu-architecture-based-on-risc-v/