Capstone Scope and Hybrid Workflow

Chapter 1 Topic 4 connects the research decisions to the concrete scope and workflow of your Autonomous Humanoid capstone.

1. What the Capstone Actually Builds

Based on the research, the capstone is a hybrid robotics/AI system that:

• Runs a humanoid (or humanoid‑like) robot in high‑fidelity simulation and, where available, on real hardware.
• Processes voice commands with Whisper and LLMs, turning natural language into structured robot intents.
• Performs perception using RGB‑D and IMU data (and optionally lidar) to detect, localize, and track objects and obstacles.
• Uses VSLAM and mapping to answer “Where am I?” and “What does my world look like?”.
• Employs Nav2‑based planning and control to move safely through complex environments.
• Executes manipulation behaviors—reaching, grasping, placing—with feedback from force/torque and joint sensors.

The end result is not a single demo script, but a reusable ROS 2 package and node graph you can extend and adapt.

2. Why “Hybrid: Simulation and Physical Deployment” Matters

The project is explicitly framed as:

Robotics/AI System (Hybrid: Simulation & Physical Deployment)

That means:

• Simulation‑only is not enough: You must design with an eye toward real sensors, friction, latency, and failures—even if you only deploy to physical hardware in final stages or shared lab time.
• Hardware‑only is unrealistic: Iterating purely on robots is too slow, risky, and expensive. Digital twins are the default environment for development and most testing.
• Interfaces must bridge both worlds: Node boundaries, topic contracts, and configuration files should support swapping between simulated and physical devices with minimal code changes.

Throughout the course you will ask, for each subsystem: How does this run in Gazebo/Isaac? How does it change when I move to the Edge Kit or a real humanoid?

3. End‑to‑End Workflow: From Research to Robot

The research phase implies a repeatable workflow you will follow at smaller scales for each feature:

1. Design and justification
   Clarify requirements and constraints, and choose tools (e.g., ROS 2 packages, Isaac extensions, message definitions) with explicit rationale.

2. Simulation‑first implementation
   Implement nodes and launch files against simulated sensors and robots. Use Gazebo/Isaac to iterate quickly on perception, planning, and control.

3. Testing and validation
   Use pytest, ROS 2 test tools, and scenario‑based simulation tests to validate behavior and performance against clear success criteria.

4. Edge deployment and tuning
   Move components that need to run on the Jetson Orin Nano or similar edge hardware, respecting its compute and memory limits.

5. Optional physical robot trials
   For labs with access to Unitree or similar platforms, run carefully staged trials, logging everything for later analysis.

This loop repeats as you add capabilities (new sensors, behaviors, or VLA skills), always grounded in the original research constraints.

4. How Scope Stays Ambitious but Achievable

The research defines a scope that is both comprehensive and bounded:

• Comprehensive because it spans the full stack: perception → mapping → planning → control → VLA.
• Bounded because:
  • You focus on one primary robot archetype (a humanoid).
  • You rely on existing libraries and stacks (ROS 2, Nav2, Isaac) instead of re‑inventing them from scratch.
  • Data storage and large‑scale infrastructure are intentionally kept out of the critical path.

In practice, this means you will:

• Own the integration and glue code that turns individual tools into a coherent humanoid system.
• Make deliberate trade‑offs when time or compute is limited, while staying within the original performance and ethical constraints.

5. How This Chapter Prepares You

By the end of Chapter 1, you should have:

• A clear picture of what you are building (the Autonomous Humanoid system).
• An understanding of where it runs (workstation, Edge Kit, optional physical robots).
• A sense of how research‑driven constraints (performance, ethics, resource limits) shape every design decision.

The remaining chapters will dive into each pillar—perception, control, simulation, and VLA—with this capstone scope and workflow as the guiding through‑line.