v3.1.0 Tensorflow Lite, RPi 4, Teachable Machine

This release updates the visual recognition approach, by running Tensorflow Lite on the RPi. The system now uses a Raspberry Pi 4, 2GB platform for much faster performance. In addition, a TRAIN action has been added to provide a UX for gathering training images from the robot camera to be used for Google's Teachable Machine to generate user trained models for deployment on the RPi.

• Non-developers should only download the delft-toolkit-3_1_0.zip file and use the /unity/delft-ai-toolkit project in Unity
• You must use the matching Raspberry Pi software and disk image available here

v3 stable release

• Robot software updated significantly - mostly fixed STT, cleaned up bugs
• Added Korean
• Added virt/phys options to many actions
• Added minimum distance to virt recognize, and made it point in the direction of the pan/tilt
• Now possible to drive arbitrary game-objects with actions by attaching ding-control-virtual component
• Bug fix for string condition w/comparison string with commas

Notes:

• For the Unity project to work out of the box, only download the delft-toolkit-3_0_3.zip file
• You must use the matching Raspberry Pi software and disk image available here

Next version we hope to include support for:

• teachable machine models
• a web hook action (e.g. to work with IFTTT, or other web services that can work with URL APIs)

minor cosmetic & bug fixes

This release provides minor fixes in interface and small bugs from the 3.0.0 release.

Download the delft-toolkit-3_0_1.zip file only

and open the

delft-toolkit-3_0_1>unity>delft-toolkit

Folder in Unity.

New hardware + other features

• Revised Hardware - The Arduino + motor shield have been eliminated and replaced by the Adafruit Raspberry Pi CRICKIT Hat. This simplifies both hardware and software.
• Changed Condition Nodes - The Float and String Condition nodes have been made easier to use by adding dropdowns for the event type
• New Sensor Capability - Added the ability to work with more than one sensor at a time, plus added the ability to use capacitive touch sensing, usable with alligator clips, copper tape, and conductive paint.
• Updated to current version of Watson - The system is updated to Watson Core SDK 1.2.0 and Watson Unity SDK 4.5
• Updated to Unity 2019.3.5 - The system now runs in the current version of Unity
• New Startup Process - The robot now speaks its IP address on boot, and here is also a new script to simplify starting the robot toolkit software
• Misc - Many small refinements and improvements have been made

Add IBM Watson Speech To Text (STT) and Text To Speech (TTS)

This release adds IBM Watson services to both Unity ("virt") and the robot ("phys"). In this release, Speech to Text and Text to Speech. We'll be adding Assistant and other Watson services moving forward.

With this update, the toolkit settings have moved from gameobjects (ding1 & ding2) in the scene to a separate central resource. Access these settings in the Unity menu: Delft AI Toolkit>Show Settings. Put your Watson iamkey for the corresponding service in the settings. This is also now where you put the IP address of the robot and the OSC marionette.

This release works with the 2019-05-16 Raspberry Pi disk image available here.

v2.0.3

• enabled action LEDs to blink on physical and virtual robots
• fixed #16 where deleting an action from an Action node with >10 actions broke the graph

v2.0.2

Changes

• Added feature that allows a variable value to be passed into the Actions node and used within action, e.g. Speak.

v2.0 Final

Final release of version 2.0. This release is the first full version of the Delft AI Toolkit 2. To use it with the physical robot, you will need to download the Raspberry Pi disk image separately.

Changes

• Added Splitter node, which routes control to one of several outlets, either randomly or sequentially
• Added AllTrue trigger to String Condition, which triggers when all triggers above are true - i.e. implements AND for triggers.

v2.0b4 Beta Release

This release is the first full version of the Delft AI Toolkit 2. To use it with the physical robot, you will need to download the Raspberry Pi disk image separately.

Changes:

• now allow multiple sensors on robot to be read
• including placeholders for user supplied audio
• started tracking RPi audio files and ML mode labels

v2.0b3 Full Release

This release is the first full release of the Delft AI Toolkit version 2. To use it with the physical robot, you will need to download the Raspberry Pi disk image separately.

Changes:

• fixed SubGraphs not exiting properly
• documentation fixes