EUP-Blockly

We present an end-user programming interface designed to allow users to author programs that instruct a simulated robot to complete tasks in a home environment. The interface uses Google's Blockly for visual programming, where users can drag and drop blocks representing robot actions, environmental states, and programming structures.

Demo

Try the interface: https://eup-blockly.netlify.app

You can experiment with different tasks and formats by modifying the URL:

• For specific tasks: https://eup-blockly.netlify.app/task{number}
• For the Full MDP format: Add ?format=FULL_MDP to the end of any URL
• For the Goal MDP format: Add ?format=GOAL_MDP to the end of any URL
• For the TAP format: Add ?format=TAP to the end of any URL
• For the Sequential format: Add ?format=SEQ to the end of any URL

Programming Paradigms

The interface supports four different end-user robot programming paradigms:

1. Sequential Programming (SEQ) - Traditional sequential programming with actions executed in order
2. Trigger-Action Programming (TAP) - Event-based programming where actions are triggered by specific conditions
3. Full MDP Programming (FULL_MDP) - Complete Markov Decision Process programming with states, actions, and goals
4. Goal-Only MDP Programming (GOAL_MDP) - Simplified MDP programming focusing only on goal states

Environment

The simulated home environment consists of four rooms:

• Kitchen
• Bedroom
• Playroom
• Porch

The robot can:

• Move between rooms
• Pick up and drop objects
• Detect presence of people and objects in the same room
• Interact with various objects (toys, coffee, mail)

Code Structure

Core Files

• scripts/obj.js - Contains all the core classes that represent objects in the game (Robot, Person, Toy)
• scripts/game.js - Contains the core game logic for robot movement and object interactions
• scripts/run.js - Handles program execution, including parsing and running user-created blocks
• scripts/blockly.min.js - The core Blockly library for visual programming

Programming Interfaces

• scripts/tap_blocks.js - Defines Blockly blocks for the TAP (Trigger-Action Programming) interface
• scripts/rl_blocks.js - Defines Blockly blocks for the RL (Reinforcement Learning) interface
• scripts/mdp.js - Handles MDP (Markov Decision Process) logic for RL format
• scripts/rl.js - Contains reinforcement learning implementation

Task-Specific Files

• scripts/settings/task{0-9}.js - Individual task configurations and settings
• task{0-9}.html - Task-specific HTML files with descriptions and UI setup

Assets & Styling

• assets/ - Contains images for:
  • Robot (robot.png)
  • Person (person.png)
  • Toys (toy1.png, toy2.png, etc.)
  • Room backgrounds
• styles/ - Contains CSS files:
  • general.css - Shared styles across tasks
  • task{1-9}.css - Task-specific styling

Interpreter

• interpreter/acorn_interpreter.js - JavaScript code interpreter for executing Blockly-generated code

Running Locally

1. Clone the repository
2. Open any of the HTML files in your browser
3. Make changes to the code and refresh the page to see updates

Block Categories

The interface provides different block categories depending on the programming paradigm:

• Actions: Robot movement and object manipulation
• States: Current conditions of the robot and environment
• Events: Trigger conditions for TAP
• Goals: Desired end states for MDP
• Controls: Programming structure blocks (conditionals, loops)

Important Notes

• The robot only perceives states within the local context (same room)
• The robot does not retain state information from previous time-steps
• We encourage running programs multiple times to ensure they work under different conditions
• Initial positions of objects and the robot may vary between runs
• The number of objects may change between runs in some tasks

Citation

If you use this project in your research, please cite:

Tewodros Ayalew*, Jennifer Wang*, Michael Littman, Blase Ur, Sarah Sebo. Enabling End Users to Program Robots Using Reinforcement Learning. HRI 2025. (* denotes equal contribution)