Developer Guide

  • 2022.3
  • 10/25/2022
  • Public Content
Contents

How it Works

The Edge Insights for Autonomous Mobile Robots (EI for AMR) modules are deployed via Docker* containers for enhanced Developer Experience (DX), support of Continuous Integration and Continuous Deployment (CI/CD) practices and flexible deployment in different execution environments, including robot, development PC, server, and cloud.
This section provides an overview of the modules and services featured with Edge Insights for Autonomous Mobile Robots.

Modules and Services

The middleware layered architecture in the Intel® oneAPI Base Toolkit (Base Kit) and Intel® Distribution of OpenVINO™ toolkit (OpenVINO™) abstracts hardware dependencies from the algorithm implementation.
The ROS 2 with data distribution service (DDS) is used as a message bus. This Publisher-Subscriber architecture based on ROS 2 topics decouples data providers from consumers.
Camera and LIDAR sensor data is abstracted with ROS 2 topics.
Video streaming processing pipelines are supported by GStreamer*. GStreamer* is a library for constructing graphs of media-handling components. It decouples sensor ingestion, video processing and AI object detection via OpenVINO™ toolkit DL Streamer framework. The applications it supports range from simple Ogg Vorbis playback audio and video streaming to complex audio (mixing) and video (non-linear editing) processing.
Also, more complex computational graphs that decouple Sense-Plan-Act autonomous mobile robot applications can be implemented using ROS 2 topic registration.
This diagram shows the software components included in the EI for AMR package. The software stack keeps evolving iteratively with additional algorithms, applications, and third-party ecosystem software components.
The EI for AMR software stack is based on software supported by and part of the underlying hardware platform, their respective Unified Extensible Firmware Interface (UEFI) based boot, and their supported Linux* operating system. For requirement details, see:

EI for AMR Drivers

Edge Insights for Autonomous Mobile Robots relies on standard Intel® Architecture Linux* drivers included and upstreamed in the Linux* kernel from kernel.org and included in Ubuntu* distributions. These drivers are not included in the EI for AMR package. Some notable drivers that are specifically important for EI for AMR include:
  • 5G/LTE Device Drivers for 5G and LTE connectivity
  • Driver for the Fibocom* FM350-G 5G/LTE modem
  • Battery Bridge Kernel Module, which allows user-space applications to feed battery and power information into the Linux kernel’s power supply subsystem. It has been designed to be used together with the ROS 2 Battery Bridge to allow ROS 2-based EI for AMR software stacks to forward battery information from an EI for AMR’s microcontroller into the Linux kernel.
  • MIPI CSI IMX390 Device Driver, for cameras that are using the Sony* IMX390 sensor and are connected to a Tiger Lake platform SoC via a MIPI CSI connection
  • Device Drivers for Intel® Movidius™ Myriad™ X VPUs
  • Video4Linux2 Driver Framework, a collection of device drivers and an API for supporting realtime video capture on Linux* systems. It supports many USB webcams, TV tuners, and related devices, standardizing their output, so that programmers can easily add video support to their applications.

EI for AMR Middleware

EI for AMR integrates the following middleware packages on EI for AMRs.
  • AAEON* ROS 2 interface, the ROS 2 driver node for AAEON EI for AMRs
  • GStreamer*, which includes support for libv4l2 video sources, GStreamer* “good” plugins for video and audio, and a GStreamer* plugin for display to display a video stream in a window
  • Kobuki, the ROS 2 driver node for Cogniteam’s Pengo robots
  • Intel® oneAPI Base Toolkit, which is able to execute Intel® oneAPI Base Toolkit sample applications. The Intel® oneAPI Base Toolkit is a core set of tools and libraries for developing high-performance, data-centric applications across diverse architectures. It features an industry-leading C++ compiler and the Data Parallel C++ (DPC++) language, an evolution of C++ for heterogeneous computing. For Intel® oneAPI Base Toolkit training and a presentation of the CUDA* converter, refer to:
  • OpenCV (Open Source Computer Vision Library), an open-source library that includes several hundred computer vision algorithms
  • Intel® Distribution of OpenVINO™ toolkit, which is a comprehensive toolkit for quickly developing applications and solutions that solve a variety of tasks including emulation of human vision, automatic speech recognition, natural language processing, recommendation systems, and many others. Based on latest generations of artificial neural networks, including Convolutional Neural Networks (CNNs), recurrent and attention-based networks, the toolkit extends computer vision and non-vision workloads across Intel hardware, maximizing performance. It accelerates applications with high-performance, AI and deep learning inference deployed from edge to cloud.
  • Intel® RealSense™ ROS 2 Wrapper node, used for Intel® RealSense™ cameras with ROS 2
  • Intel® RealSense™ SDK, used to implement software for Intel® RealSense™ cameras
  • ROS 2 ros1_bridge, which provides a network bridge allowing the exchange of messages between ROS1 and ROS 2. This lets users evaluate the EI for AMR SDK on EI for AMRs or with sensors for which only ROS1 driver nodes exist.
  • ROS 2, Robot Operating System (ROS), which is a set of open source software libraries and tools for building robot applications
    ROS 2 depends on other middleware, like the Object Management Group (OMG) DDS connectivity framework that is using a publish-subscribe pattern. The standard ROS 2 distribution includes eProsima Fast DDS implementation.
  • ROS 2 Battery Bridge, which utilizes the Battery Bridge Kernel Module to forward battery information from an EI for AMR’s microcontroller into the Linux kernel
  • RPLIDAR ROS 2 Wrapper node, for using RPLIDAR LIDAR sensors with ROS 2
  • SICK Safetyscanners ROS 2 Driver, which reads the raw data from the SICK Safety Scanners and publishes the data as a laser_scan msg
  • Teleop Twist Joy, a generic facility for teleoperating twist-based ROS 2 robots with a standard joystick. It converts joy messages to velocity commands. This node provides no rate limiting or autorepeat functionality. It is expected that you take advantage of the features built into ROS 2 Driver for Generic Joysticks for this.
  • Teleop Twist Keyboard generic keyboard teleoperation for ROS 2
  • Twist Multiplexer for when there is more than one source to move a robot with a geometry_msgs::Twist message. It is important to multiplex all input sources into a single source that goes to the EI for AMR control node.
  • ROS 2 Driver for Generic Joysticks
  • ModemManager, which provides a unified high level API for communicating with mobile broadband modems, regardless of the protocol used to communicate with the actual device (for example, generic AT, vendor-specific AT, QCDM, QMI, MBIM). Edge Insights for Autonomous Mobile Robots uses ModemManager to establish 5G connections using the Fibocom* FM350-G 5G/LTE modem.

EI for AMR Algorithms

EI for AMR includes reference algorithms as well as deep learning models as working examples for the following automated robot control functional areas:
  • DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is an unsupervised clustering algorithm that clusters high dimensional points based on their distribution density. Adaptive DBSCAN (ADBSCAN) has clustering parameters that are adaptive based on range and are especially suitable for processing LIDAR data. It improves the object detection range by 20-30% on average.
  • Collaborative visual SLAM, a collaborative visual simultaneous localization and mapping (SLAM) framework for service robots. With an edge server maintaining a map database and performing global optimization, each robot can register to an existing map, update the map, or build new maps, all with a unified interface and low computation and memory cost. A collaborative visual SLAM system consists of at least two elements:
    • The
      tracker
      is a visual SLAM system with support for inertial and odometry input. It estimates the camera pose in real-time, and maintains a local map. It can work without a server, but if it has one configured, it communicates with the server to query and update the map. The tracker represents a robot. There can be multiple trackers running at the same time.
    • The
      server
      maintains the maps and communicates with all trackers. For each new keyframe from a tracker, it detects possible loops, both intra-map and inter-map. Once detected, the server performs map optimization or map merging and distributes the updated map to corresponding trackers.
    For collaborative visual SLAM details, refer to A Collaborative Visual SLAM Framework for Service Robots paper.
  • FastMapping, which is an algorithm to create a 3D voxel map of a robot’s surrounding, based on Intel® RealSense™ depth sensor data.
  • OpenVINO™ Model Zoo, optimized deep learning models and a set of demos to expedite development of high-performance deep learning inference applications. A developer can use these pre-trained models instead of training their own models to speed-up the development and production deployment process.
  • ROS 2 Cartographer, a system that provides real-time simultaneous localization and mapping (SLAM) based on real-time 2D LIDAR sensor data. It is used to generate as-built floor plans in the form of occupancy grids.
  • ROS 2 Depth Image to Laser Scan, which converts a depth image to a laser scan for use with navigation and localization.
  • ROS 2 Navigation stack, which seeks a safe way to have a mobile robot move from point A to point B. This completes dynamic path planning, computes velocities for motors, detects and avoids obstacles, and structures recovery behaviors. Navigation 2 uses behavior trees to call modular servers to complete an action. An action can be computing a path, controlling effort, recovery, or any other navigation-related action. These are separate nodes that communicate with the behavior tree over a ROS 2 action server.
  • RTAB-Map (Real-Time Appearance-Based Mapping), a RGB-D, Stereo and Lidar Graph-Based SLAM approach based on an incremental appearance-based loop closure detector. The loop closure detector uses a bag-of-words approach to determinate how likely a new image comes from a previous location or a new location. When a loop closure hypothesis is accepted, a new constraint is added to the map’s graph, then a graph optimizer minimizes the errors in the map. A memory management approach is used to limit the number of locations used for loop closure detection and graph optimization, so that real-time constraints on large-scale environnements are always respected. RTAB-Map can be used alone with a handheld Kinect, a stereo camera or a 3D lidar for 6DoF mapping, or on a robot equipped with a laser rangefinder for 3DoF mapping.
  • IMU Tools - filters and visualizers - from https://github.com/CCNYRoboticsLab/imu_tools:
    • imu_filter_madgwick: A filter which fuses angular velocities, accelerations, and (optionally) magnetic readings from a generic IMU device into an orientation.
    • imu_complementary_filter: A filter which fuses angular velocities, accelerations and (optionally) magnetic readings from a generic IMU device into an orientation quaternion using a novel approach based on a complementary fusion.
    • rviz_imu_plugin: A plugin for rviz which displays sensor_msgs::Imu messages.
  • Intelligent Sampling and Two-Way Search (ITS) global path planner Robot Operating System 2 (ROS 2) Plugin is a plugin for ROS 2 Navigation package which conducts a path planning search on a roadmap from two directions simultaneously. The main inputs are 2D occupancy grid map, robot position, and the goal position. The occupancy is converted into a roadmap and can be saved for future inquiries. The output is a list of waypoints which constructs the global path. All inputs and outputs are in standard ROS 2 formats. This plugin is a global path planner module which is based on the Intelligent Sampling and Two-Way Search (ITS). Currently, the ITS plugin does not support continuous replanning. To use this plugin, a simple behavior tree with compute path to pose and follow path should be used. The inputs for the ITS planner are global 2d_costmap (nav2_costmap_2d::Costmap2D), start and goal pose (geometry_msgs::msg::PoseStamped). The outputs are 2D waypoints of the path. The ITS planner gets the 2d_costmap and it converts it to either Probabilistic Road Map (PRM) or Deterministic Road Map (DRM). The generated roadmap is saved in a txt file which can be reused for multiple inquiries. Once a roadmap is generated, the ITS conducts a two-way search to find a path from the source to destination. Either the smoothing filter or catmull spline interpolation can be used to create a smooth and continuous path. The generated smooth path is in the form of ROS navigation message type (nav_msgs::msg).
  • Kudan Visual SLAM (KdVisual), Kudan’s proprietary visual SLAM software, has been extensively developed and tested for use in commercial settings. Open source and other commercial algorithms struggle in many common use cases and scenarios. Kudan Visual SLAM achieves much faster processing time, higher accuracy, and a more robust results in dynamic situations.
  • The Point Cloud Library (PCL), a standalone, large scale, open project for 2D/3D image and point cloud processing (see also https://pointclouds.org/). The EI for AMR SDK version of PCL adds optimized implementations of several PCL modules which allow you to offload computation to a GPU.
  • Robot_localization (from https://github.com/cra-ros-pkg/robot_localization), a collection of state estimation nodes, each of which is an implementation of a nonlinear state estimator for robots moving in 3D space. It contains two state estimation nodes, ekf_localization_node and ukf_localization_node. In addition, robot_localization provides navsat_transform_node, which aids in the integration of GPS data.
  • SLAM Toolbox, a set of tools and capabilities for 2D SLAM built by Steve Macenski that includes the following.
    • Starting, mapping, saving pgm files, and saving maps for 2D SLAM mobile robotics
    • Refining, remapping, or continue mapping a saved (serialized) pose-graph at any time
    • Loading a saved pose-graph continue mapping in a space while also removing extraneous information from newly added scans (life-long mapping)
    • An optimization-based localization mode built on the pose-graph. Optionally run localization mode without a prior map for “LIDAR odometry” mode with local loop closures
    • Synchronous and asynchronous modes of mapping
    • Kinematic map merging (with an elastic graph manipulation merging technique in the works)
    • Plugin-based optimization solvers with an optimized Google* Ceres-based plugin
    • rviz2 plugin for interacting with the tools
    • Graph manipulation tools in rviz2 to manipulate nodes and connections during mapping
    • Map serialization and lossless data storage

EI for AMR Applications

  • Intel In-Band Manageability, monitors device(s) and updates software and firmware of the device(s) remotely.
  • Object Detection AI Application, detects objects in video data using a deep learning neural network model from the OpenVINO™ Model Zoo.
  • VDA5050 Sample Handler, processes selected commands from the VDA5050 EI for AMR/AGV interoperability standard and forwards the EI for AMR’s software components for autonomous navigation.
  • Wandering Application, included in the EI for AMR SDK to demonstrate the combination of the middleware, algorithms, and the ROS 2 navigation stack to move a robot around a room avoiding hitting obstacles, updating a local map in real time exposed as ROS topic, and publish AI-based objects detected in another ROS topic. It uses the robot’s sensors and actuators that are available from the robot’s hardware configuration.

Edge Server Middleware

  • FIDO Device Onboarding, an automatic onboarding protocol for IoT devices. Permits late binding of device credentials, so that one manufactured device may onboard, without modification, to many different IoT platforms.
  • Intel® Smart Edge Open, a software toolkit for building edge platforms. It speeds up development of edge solutions that host network functions alongside AI, media processing, and security workloads with reference solutions optimized for common use cases powered by a Certified Kubernetes* cloud native stack.
  • kube-proxy, the Kubernetes* network proxy that runs on each node. This reflects services as defined in the Kubernetes* API on each node and can do simple TCP, UDP, and SCTP stream forwarding or round robin TCP, UDP, and SCTP forwarding across a set of backends. Service cluster IPs and ports are currently found through Docker* links-compatible environment variables specifying ports opened by the service proxy. There is an optional addon that provides cluster DNS for these cluster IPs. The user must create a service with the apiserver API to configure the proxy.
  • The VDA5050 ROS 2 Bridge which translates VDA5050 messages into ROS 2 messages which can be received and executed by ROS 2 components in an EI for AMR-SDK based EI for AMR control system.
  • The VDA Navigator, a basic waypoint navigator that receives a list of waypoints and sends them to the ROS 2 Navigation stack one after the other

Edge Server Algorithms

  • OpenVINO™ Model Zoo, which includes optimized deep learning models and a set of demos to expedite development of high-performance deep learning inference applications. A developer can use these pre-trained models instead of training their own models to speed-up the development and production deployment process.

Edge Server Applications

  • OpenVINO™ Model Server (OVMS), a high-performance system for serving machine learning models. It is based on C++ for high scalability and optimized for Intel solutions, so that you can take advantage of the power of the Intel® Xeon® processor or Intel’s AI accelerator and expose it over a network interface. OVMS uses the same architecture and API as TensorFlow Serving, while applying OpenVINO for inference execution. Inference service is provided via gRPC or REST API, making it easy to deploy new algorithms and AI experiments.
  • ThingsBoard*, an open-source IoT platform for data collection, processing, visualization, and device management. It enables device connectivity via industry standard IoT protocols - MQTT, CoAP and HTTP and supports both cloud and on-premises deployments.

Tools

ROS Tools
Edge Insights for Autonomous Mobile Robots is validated using ROS 2 nodes. ROS 1 is not compatible with EI for AMR components. A ROS 1 bridge is included to allow EI for AMR components to interface with ROS 1 components.
  • From the hardware perspective of the supported platforms, there are no known limitations for ROS 1 components.
  • For information on porting ROS 1 applications to ROS 2, here is a guide from the ROS community.
Edge Insights for Autonomous Mobile Robots includes:
  • colcon (collective construction), a command line tool to improve the workflow of building, testing, and using multiple software packages. It automates the process, handles the ordering, and sets up the environment to use the packages.
  • rqt, a software framework of ROS 2 that implements the various GUI tools in the form of plugins.
  • rviz2, a tool used to visualize ROS 2 topics.
Simulation
Edge Insights for Autonomous Mobile Robots includes:
  • The Gazebo* robot simulator, making it possible to rapidly test algorithms, design robots, perform regression testing, and train AI systems using realistic scenarios. Gazebo offers the ability to simulate populations of robots accurately and efficiently in complex indoor and outdoor environments.
  • An industrial simulation room model for Gazebo*, the Open Source Robotics Foundation (OSRF) Gazebo Environment for Agile Robotics (GEAR) workcell that was used for the ARIAC competition in 2018.
Other Tools
Edge Insights for Autonomous Mobile Robots includes:
  • Intel® oneAPI Base Toolkit, which includes the DPC++ compiler and compatibility tool, as well as debugging and profiling tools like VTune™ Profiler, etc. (formerly known as Intel System Studio).
  • OpenVINO™ Tools, including the model optimization tool.

Deployment

All applications, algorithms, and middleware components which are executed as standalone processes are deployed in their own Docker* containers. This allows you to selectively pull these components onto an EI for AMR or Edge Server and launch them there.
For development purposes, the middleware libraries and all tools are deployed in a single container called
Full SDK
. This container is constructed hierarchically by extending the
OpenVINO SDK
container, which itself extends the
ROS2 SDK
container. For storage space savings, you can choose to run any of the containers depending on the needs of your application.
  • The
    ROS2 SDK
    container includes the ROS 2 middleware and tools, Intel® RealSense™ SDK and ROS 2 wrapper, GStreamer* and build tools, ROS 2 packages (Cartographer, Navigation, RTAB_MAP) and the FastMapping application (the Intel-optimized version of octomap).
  • The
    OpenVINO SDK
    container includes the
    ROS2 SDK
    , as well as the OpenVINO™ development toolkit, the OpenVINO™ DL GStreamer* plugins and the Wandering demonstration application.
  • The
    Full SDK
    container includes the OpenVINO™ container, as well as the Intel® oneAPI Base Toolkit, the Data Parallel C++ (DPC++) compatibility tool and profiler, analyzer tools.

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.