Topic 5 — Development Workflow, Storage Model & Testing

This topic ties together three key aspects from the course research:

• How you develop and iterate on humanoid behaviors,
• Why the system uses a minimal storage model, and
• How you test and validate your work using ROS 2 tools and pytest.

---

1. Development Workflow: From Idea to Robot Behavior

Your typical workflow in this course looks like:

1. Design the behavior

   • Clarify the user story (“Robot, pick up the cup and place it on the table”).
   • Identify which pillars are involved (perception, planning, control, VLA).

2. Prototype in Python + ROS 2

   • Implement initial logic in Python 3.x, using ROS 2 nodes for I/O.
   • Keep nodes small and focused (e.g., one for perception, one for planning, one for control).

3. Integrate with Simulation

   • Connect nodes to Gazebo or Isaac Sim for end‑to‑end tests in a digital twin environment.
   • Iterate on algorithms and parameters until behavior is stable in simulation.

4. Refine for Performance & Constraints

   • Profile latency and resource use.
   • Simplify models or adjust rates where necessary to meet real‑time goals.

5. Deploy to Edge (Jetson) and Validate

   • Move the validated subset of nodes and models to the edge kit.
   • Re‑run critical tests with real sensors where available.

Throughout this process, you are expected to:

• Use Spec‑Driven tasks and specs in specs/1-physical-ai-course/ as your guide.
• Keep ROS 2 interfaces consistent between simulation and hardware.
• Document non‑obvious decisions in your project notes or documentation.

---

2. Storage Model: Real-Time Focus, Minimal Databases

The research recommends no dedicated persistent database inside the core robotics stack:

• Sensor streams (RGB‑D, LiDAR, IMU, audio) and control commands are ephemeral.
• Real‑time performance is the priority; disk or network latency should not be on the critical path.

Instead:

• Use ROS 2 logging and bag files for recording experiments and debugging.
• Store training data, evaluation results, or long‑term logs in files or external systems.
• Introduce databases only when they provide clear value (e.g., long‑term telemetry in a separate project).

This matches the project’s emphasis on real‑time interaction with the physical world, not on building a generic data warehouse.

---

3. Testing Strategy: ROS 2 Test Tools + pytest

Testing spans multiple levels:

3.1 Unit Tests with pytest

Use pytest to validate:

• Pure Python functions (e.g., planners, filters, cost functions).
• Utility modules that transform or validate data before it reaches ROS 2.

Benefits:

• Fast feedback loops.
• Easy to run locally and in CI.
• Encourages clean separation between “logic” and “ROS 2 glue code”.

3.2 Integration & System Tests with ROS 2 Tools

Use ROS 2’s native tools for end‑to‑end testing:

• rostest / gtest for:
  • Verifying that publishers and subscribers communicate correctly.
  • Ensuring actions and services behave as expected.
  • Testing multi‑node scenarios (e.g., perception → planning → control).

Examples:

• A test that brings up a minimal navigation stack, sends a goal, and asserts that a simulated robot reaches the target region.
• A test that runs a perception node on recorded RealSense data and verifies detections.

---

4. Quality & Safety in the Loop

Because this course is about Physical AI, quality and safety are integral to testing:

• Tests should confirm not only correct outputs, but also:
  • Reasonable execution times (no unbounded delays in control loops).
  • Safe behavior when inputs are missing, noisy, or adversarial.
• When designing labs and the capstone:
  • Prefer behaviors that fail safely (e.g., stop motion, enter a safe pose) over ones that silently misbehave.
  • Treat Constitution principles (especially around safety and beneficial outcomes) as part of the definition of “done”.

---

5. How This Informs Your Capstone Work

By the time you reach the capstone, you should be comfortable with:

• Iterating quickly using Python + ROS 2 and simulation.
• Using minimal, focused storage solutions appropriate for real‑time robotics.
• Writing both pytest and ROS 2‑based tests to validate behavior.

These habits will make your final autonomous humanoid project:

• Easier to debug and reason about.
• More robust under real‑world noise and constraints.
• Better aligned with how modern robotics teams actually build, test, and ship complex systems.