Skip to content

ADLINK tegraBot

People-tracking robot according to cloth type using AI

source code is here

About the project

The purpose of this pkg is to demonstrate the abilities of ADLINK M200-JT2 computing platform. * ADLINK M200-JT2 Edge Inference Platform: with NVIDIA® Jetson™ TX2 [link]

Dev Team

  • Ewing Kang (ewing.kang@adlinktech.com)
  • Alan Chen (alan.chen@adlinktech.com)
  • Chester Tseng (chester.Tseng@adlinktech.com)
  • Bill Wang (bill.wang@adlinktech.com)
  • Ryan Chen (ryanjb.chen@adlinktech.com)

ADLINK Technology, Inc
Advanced Robotic Platform Group

License

Apache 2.0
Copyright 2018 ADLINK Technology, Inc.

Tutorial

Prerequisites

Hardware

  • Depth camera: Realsense D400 / Astra / Astra Pro
    Source: realsense, astra ros
    Binary for Astra: $ sudo apt-get install ros-kinetic-astra*
    Notice:

    1. About RealSense SDK 2.0, we highly recommed binary version.
    2. Remember to create udev. If possible, please buy Astra instead of Astra Pro! Testing: $ roslaunch realsense2_camera demo_pointcloud.launch

  • YDLidar
    Source: https://github.com/EAIBOT/ydlidar
    Notice: Remember to laod udev. Could be replaced by any type of lidar.
    Testing: $ roslaunch ydlidar x4.launch

ROS

  • Navigation
    Installation: $ sudo apt-get install ros-kinetic-navigation*
    Source: https://github.com/ros-planning/navigation
    Notice: if "replan" mode of global planner is malfunctioned, please compile whole pkgs from source.
    Testing: $ roslaunch spencer_people_tracking_launch tracking_on_bagfile.launch

  • Turtlebot2
    Installation: $ sudo apt-get install ros-kinetic-turtlebot
    Source: https://github.com/turtlebot/turtlebot

AI tracking

You need these package installed for the tensorflow_object_detector node to work CUDA v8.0 cudnn v6.0 libuvc (training only) protoc 3.3 (notice the version)

There are 4 packages working together that made the people tracking using AI possible. 

cd ~/catkin_ws/src

git clone https://github.com/Adlink-ROS/adlink_tegrabot https://github.com/Adlink-ROS/tf_ai_tracker.git https://github.com/Adlink-ROS/tegrabot_description.git https://github.com/Adlink-ROS/tensorflow_object_detector.git
  • adlink_tegrabot: main launch collection of the tegrabot
  • tegrabot_description: robot model (urdf) for rviz
  • tensorflow_object_detector: TensorFlow classifier
  • tf_ai_tracker: people tracking controller

Running the demo

Before your first run

  • (optional) Mapping & Time Synchronizing

Host launch

  • ROS: $ roslaunch adlink_tegrabot NeuronBot_Demo_Host_AIO.launch
  • shell script: `$ ./PATH_TO_WORKSPACE/adlink_tegrabot/autostart/NeuronBot_Demo_Host_AutoStart.sh)

visualization

rviz file neuronbot_demo_client.rviz is in the rviz folder

Optional ROS package

  • SPENCER
    Binary: https://github.com/spencer-project/spencer_people_tracking#installation-from-l-cas-package-repository
    Source: https://github.com/spencer-project/spencer_people_tracking#installation-from-source
    Notice: Unless you want to use HOG+SVM [CUDA required], we highly recommend binary version.

  • leg_tracker
    Source: https://github.com/angusleigh/leg_tracker
    Notice: The kinetic branch only supports OpenCv 3.3 and higher ver.
    Testing: $ roslaunch leg_tracker demo_stationary_simple_environment.launch

Training your own model

Please check our wiki for more information.

Project status

v1.0

Roadmap

  • [ ] Multi robot example