KAI The German Crane AI

Artificial Intelligence for crane steering to replace human operator

The Crane Lab

Demonstration 2:13 min

Crane AI Components and Cognitive Architecture

Sensors
Sensor Nodes
World model
Decision Making
Cloud Infrastructure

Perceive the environment with sensors.

  • Onboard LiDAR continuously provides 3D scans of the inventory and environment
  • Cameras detect the position of the grapple, load, and other objects
  • Optional recording of reference points with a dense drone scan

    • Sensors

    Recognize load, grapple, and trucks.

    Software detectors on specialized hardware modules recognize the objects that matter and provide critical information to the AI in real-time.

    Sensor Nodes

    Populate and maintain 
the world model.

    The world model is the continuously updated representation of the AI’s beliefs about the world.

    Information from sensor nodes is fused into one unified representation of the work area.

    The positions of relevant objects in the work area are stored, and the position of the crane and its moving parts is estimated and continuously updated in the world model.

    World model

    Deliberation, strategies, and behaviors.

    The Crane AI is constantly making high and low-level decisions. What is the next step in the process? Where to place the load? Which is the most efficient action to achieve the present goal? The decisions are reevaluated every control cycle as updated information from the sensors becomes available. The AI reacts immediately to changes in the environment or to external requests.

    Collect, store, and index petabytes of data for later use by the AI.

    PSIORI’s cloud Infrastructure can collect data from various sources and store them safely using a cellular connection or an existing internet connection to our managed cloud services or to a private instance in AWS or Azure. An automated index allows retrieval of pictures with specific content, for instance “pictures with open grapple”. This allows the close monitoring and troubleshooting of the crane remotely and also enables continuous learning: the ability to refine and improve the AI’s models and decisions according to the data collected during operation.

    Whereas the cloud connection eases the deployment and customization of the Crane AI, it is purely optional in operation. An on-premises alternative can also be provided.

    Cloud Infrastructure

    Why is artificial intelligence more effective, safer, more reliable, and more economical than humans?

    Read the article

    AI Features

    Crane Driver Assistance
    Movement Control
    Sway Control
    Modern HMI
    Reports

    Provide crane operator with assistance systems that help him do a better job, avoid mistakes, or train for special occurrences.

    PSIORI’s Crane Assist warns operators if a dangerous situation occurs, offers hints for better performance, or even trains an operator by issuing commands and taking over if the situation gets out of control.

    Crane Driver Assistance

    Components needed: AI Injector, Sway Control, HMI.

    KAI can send commands to the PLC and steer the crane. A human operator can take over command and steer the crane with one click.

    Movement Control

    Components needed: Decisions making Module, AI Injector

    KAI can avoid sway or dampen existing sway and restore controlled movement.

    Large masses at the end of a crane’s grapple become serious and dangerous problems when they sway uncontrollably. Sway can occur either from unbalanced joystick movement or through external influences like wind or collisions. Sway can be controlled and cancelled on all movement axes, including rotational sway on the load.

    Components needed: AI infusion, Sway Control

    Provides essential information about the crane AI’s planned actions and serves as remote control for AI components.

    PSIORI’s iPad-based Human Machine Interface allows an operator to oversee the crane’s functions and to turn on or off the crane’s autonomy. An optional dead man’s switch function ensures that the operator is always within reach of the HMI and able to stop the crane. Everything is presented in a clean and modern interface.

    Modern HMI

    Components needed: AI Injector, HMI, Cloud infrastructure

    Provide the plant managers with KPIs about the crane’s performance.

    PSIORI’s shift log describes the crane’s activities during the last shift or any other arbitrary period. It is provided as a PDF or via a web-based tool (supervisor UI). It allows management to view the crane’s performance over time and evaluate performance on a day-by-day (or shift-by-shift) basis. Crane performance metrics include among others tons of logs shifted to deck and stack, net inventory change, number of pick-up and deposit events, movement totals by axis, and even the change in the shape of the log pile.

    The supervisor UI is a mobile phone-compatible web-based dashboard for the plant supervisor and management that shows selected key performance indicators (shifted weight, number of deposit events...) during any given period (shift, day, week...). It also shows a live video from the crane’s cameras.


    Reports