NUsight: Building an
effective debugging tool

Josephus Paye II

Co-Team Leader, NUbots

NUsight

A collection of tools that provide realtime visualisation of debug data from one or more robots

🌐 web-based

many 🤖 to many 💻

The Elements of a Visual Debugging Tool

The Elements of a Visual Debugging Tool

Data

Rendering

Networking

the values and images to visualise

getting data from a robot into the debugging tool

displaying the data in a user interface, with interactive controls

Five Key Principles

Five Key Principles

Keep it simple

Use UDP with stateless messages

Use declarative, data-driven rendering

Go beyond numbers and graphs

Give users control over what is visualised

1. Keep It Simple

1. Keep It Simple

The system

is wrong

The debugger

is wrong

1. Keep It Simple

Minimise the number of settings and configuration options in the tool

Adopt design patterns and principles
that exclude certain classes of bugs

Keep the tool strictly as a viewer, responsible for displaying data only

2. Use UDP with Stateless Messages

2. Use UDP with Stateless Messages

TCP vs UDP

2. Use UDP with Stateless Messages

message Image {
    message Lens {
        enum Projection {
            UNKNOWN     = 0;
            RECTILINEAR = 1;
            EQUIDISTANT = 2;
            EQUISOLID   = 3;
        }

        Projection projection   = 1;
        float      focal_length = 2;
        float      fov          = 3;
        fvec2      centre       = 4;
        fvec2      k            = 5;
    }

    uint32                    format     = 1;
    uvec2                     dimensions = 2;
    bytes                     data       = 3;
    uint32                    camera_id  = 4;
    string                    name       = 5;
    google.protobuf.Timestamp timestamp  = 6;
    mat4                      Hcw        = 7;
    Lens                      lens       = 8;
}

Stateless messaging

3. Use Declarative,
Data-driven Rendering

3. Use Declarative,
Data-driven Rendering

Declarative vs Imperative Rendering

<!-- Declarative: SVG -->
<svg width="200" height="200">
    <rect width="100" height="100" fill="green" />
</svg>

<!-- Imperative: Canvas 2D -->
<canvas id="canvas" width="200" height="200" />
<script>
    let canvas = document.getElementById('canvas');
    let ctx = canvas.getContext('2d');
    ctx.fillStyle = 'green';
    ctx.fillRect(0, 0, 100, 100);
</script>

3. Use Declarative,
Data-driven Rendering

Data-driven Rendering

React.js

MobX

4. Go Beyond Numbers and Graphs

4. Go Beyond Numbers and Graphs

Make it look real where possible

Localisation

Vision

4. Go Beyond Numbers and Graphs

Visualise with context and system info

5. Give Users Control Over What is Visualised

5. Give Users Control Over What is Visualised

Give control where the data goes in (input)

# NUsight.yaml

# Select which data to send
reaction_handles:
  # Overview information for the dashboard, designed for minimum bandwidth
  overview: true
  # Motor positions, orientation, etc.
  sensor_data: false
  # Raw image
  image: false
  # compressed image
  compressed_image: false
  # Classified image
  classified_image: false
  # Vision objects including goals, balls and debugging lines
  vision_object: false
  # Visual mesh data
  visual_mesh: false
  # Green horizon data
  green_horizon: false
  # Localisation state estimates and uncertainties
  localisation: false
  # Data for chart tool in NUsight
  data_point: false

5. Give Users Control Over What is Visualised

Give control where the data comes out

Conclusion

Keep it simple

Use UDP with stateless networking

Use declarative, data-driven rendering

Go beyond numbers and graphs

Give users control over what is visualised

A realtime visual debugging tool is very useful and improves development efficiency when building complex systems like soccer playing robots.

Acknowledgement

Thanks to Brendan Annable, creator of NUsight, and Trent Houliston, contributor to NUsight, as well as the rest of the NUbots team for their help in preparing this presentation.

Links

NUsight is open source and available on GitHub (github.com/NUbots/NUsight2) and can be referred to when building your own debugging tool.

We are also happy to answer any questions you may have - email nubots@newcastle.edu.au

Thanks!

NUsight: Building an effective debugging tool for soccer playing robots

By josephuspaye

NUsight: Building an effective debugging tool for soccer playing robots

This talk was given remotely at the Humanoid Virtual Workshops held for RoboCup 2020. More information about NUsight and NUbots can be found at https://nubook.nubots.net

  • 335
Loading comments...