01 /
GENERATE
Describe the task
Write what you want the robot to do in plain language. Muse Spark turns the task into objectives and a ROS 2 Python controller.
human intent controller.py
AI × Robotics × Simulation
Make AI-generated
robot behaviors
prove they work.
HeyWalk turns human intent into executable robot behavior. It then runs that behavior in a physics-based simulation, measures what happened, and uses the evidence to improve.
Platform overview
Task → Controller → MuJoCo Simulation → Measured Verdict
Simulated robot: MicroDuck by Pollen Robotics (CC BY-NC-SA 4.0).
Powered by open-source robotics & AI
Muse AI
MuJoCo
ONNX
ROS 2
MicroDuck
NumPy
Python
FastAPI
React
Three.js
Hugging Face
The HeyWalk loop
From thought to motion.
With evidence at every step.
A complete experiment cycle connects AI-generated control logic to observable robot behavior.
02 /
EXECUTE
Run in simulation
The controller runs in an isolated container and communicates through ROS 2 topics. MuJoCo advances the robot's physics.
ROS 2 MuJoCo
03 /
VERIFY
Measure what happened
An independent evaluator checks telemetry against task and safety criteria. The model cannot declare its own success.
telemetry verdict
04 /
IMPROVE
Diagnose and rerun
When a run fails, Muse Spark receives the evidence, proposes a targeted patch, and the updated controller is tested again.
evidence patch
Each run preserves the task, controller version, environment seed, logs, telemetry, evaluator verdict, and
correction history. Reruns can use the same world configuration for meaningful comparison.
See the experiment
Watch HeyWalk in action.
We ask the MicroDuck to walk to the green flag then back to starting point to stop safely.
37-second simulation walkthrough · Simulated robot: MicroDuck by Pollen Robotics (CC BY-NC-SA 4.0)
A call to Indonesian campuses & vocational high schools
The next generation of robotics talent starts before the industry arrives.
Indonesia's students should not have to wait for expensive robot hardware to start learning robotics. With simulation-first experimentation, campuses and vocational high schools can introduce robot programming, control logic, AI-assisted development, and evidence-based testing using accessible computing resources.
We are looking for universities, polytechnics, robotics clubs, lecturers, and vocational high school programs interested in exploring a hands-on HeyWalk pilot. The goal is to help students build practical skills for a future where robotics becomes part of more industries.
The pilot inquiry opens a pre-filled GitHub issue. No external form backend is required.
For universities
Bring AI-assisted robotics experiments into robotics, mechatronics, computer science, and engineering learning environments.
For vocational high schools
Introduce students to programming, automation concepts, simulation, and the foundations of robotics without starting with a costly robot fleet.
For robotics clubs
Give student teams a shared environment to prototype behaviors, test ideas, inspect failures, and compare iterations.
For industry partners
Help shape practical learning pathways and expose emerging talent to the tools and thinking behind robotics development.
The invitation
Let's make robotics more accessible in Indonesia.
Start with a simulation. Build the skills. Be ready for the opportunities ahead.
Documentation & getting started
Go from curious
to running an
experiment.
HeyWalk is a software-in-the-loop research prototype. Explore the repository for setup instructions, implementation details, API routes, safety boundaries, and third-party notices.
Quick start
Install the stack, configure a Meta Model API key, start the app, and open the workspace locally.
curl
-fsSL
https://raw.githubusercontent.com/wansatya/heywalk/main/install.sh | bash
How a run works
Understand task parsing, controller generation, isolated execution, ROS 2 topics, telemetry, evaluation, and correction.
Generate → Execute → Evaluate → Fix
Simulation & policy
Learn how MuJoCo handles physics while the upstream ONNX policy provides the MicroDuck's balance and gait.
MuJoCo + ONNX Runtime + MicroDuck
Licenses & attribution
Review the separate licenses for HeyWalk code, MicroDuck MJCF, STL meshes, pretrained policy, and runtime dependencies.
MIT · Apache-2.0 · CC BY-NC-SA 4.0
Important: simulation is not physical-world certification.
HeyWalk currently targets software-in-the-loop experiments. A successful simulation does not prove physical robot safety. Generated code is untrusted and runs in an isolated, resource-limited container. Do not connect this prototype to a physical robot.