Humanoid WBC Thesis

Development and Simulation of a Whole-Body Control Framework for a Humanoid Robot Master's Thesis — David — 2026

---

Research Question

How can a whole-body control framework be designed and validated in simulation to enable a humanoid robot to perform stable, multi-contact locomotion and manipulation tasks under dynamic constraints?

---

Overview

This repository organizes the full research workflow of the Master's thesis, from problem definition to publication-grade manuscript. The work proposes a hierarchical WBC architecture: a centroidal MPC planning layer coupled with a QP-based Quadratic Program execution layer, validated across five simulation scenarios (S1–S5) on a humanoid URDF model.

The research pipeline combines a multi-LLM workflow (Perplexity, Gemini, Claude) with adversarial and council analysis, targeted gap-filling, simulation development, and structured academic writing aimed at publication-level quality.

---

Thesis Chapters

Chapter	Title	Status
1	Introduction	🔲 Not started
2	State of the Art — WBC for Humanoids	🔲 Not started
3	Methodology	🔲 Not started
4	Simulation Framework	🔲 Not started
5	Experiments and Results	🔲 Not started
6	Discussion	🔲 Not started
7	Conclusion and Future Work	🔲 Not started

---

Technical Stack (Planned)

Component	Tool
Dynamics library	Pinocchio
Optimal control	Crocoddyl
QP solver	OSQP / qpOASES
Simulator	MuJoCo or NVIDIA Isaac Sim (TBD)
Robot model	TALOS or Unitree H1 URDF (TBD)
Language	Python / C++

---

Test Scenarios

ID	Description	Primary Metric
S1	Static balancing under external pushes	CoM error, recovery time
S2	Flat-ground walking (3 m)	Task success, CoM tracking
S3	Stair climbing (3 steps)	Foot clearance, contact forces
S4	Loco-manipulation (walk + reach)	End-effector error + CoM tracking
S5	Perturbation robustness	Recovery rate

---

Repository Structure

humanoid-wbc-thesis/
├── 00_problem_definition/     # Research question, scope boundaries, hypotheses
├── 01_prompts/                # Canonical LLM prompts (versioned for reproducibility)
├── 02_raw_research/           # Raw LLM outputs, PDFs — read-only once archived
├── 03_processed/              # Summaries, extracted claims, taxonomy, knowledge graph
├── 04_council_analysis/       # Contradictions, consensus, gaps, confidence scores
├── 05_targeted_research/      # Second-pass gap-filling research
├── 06_writing/                # Outline → drafts → final chapters
├── 07_review/                 # Adversarial review, supervisor feedback, revisions
├── 08_references/             # BibTeX, validated sources, Zotero exports
├── 09_appendices/             # Frozen code, figures, datasets for manuscript
├── 10_simulation/             # URDF models, simulator configs, controller source
├── 11_experiments/            # Experiment logs, plots, tabulated results
├── 12_meetings/               # Supervisor meeting notes and action items
├── 13_logbook/                # Daily research journal
└── master_memory.md           # Single source of truth — read before any work session

---

Research Workflow

1. Define (00_) — research question, scope, hypotheses
2. Prompt (01_) — author and version LLM prompts
3. Collect (02_) — run LLMs, archive raw outputs and primary PDFs
4. Process (03_) — summarize, extract atomic claims, build taxonomy
5. Analyze (04_) — council/adversarial cross-examination; flag contradictions and gaps
6. Refine (05_) — targeted research to close gaps
7. Simulate (10_, 11_) — build/modify URDF, run controllers, record experiments
8. Write (06_) — outline then chapter drafts
9. Review (07_) — adversarial self-review + supervisor loop
10. Finalize (08_, 09_) — consolidate references and appendices

---

Git Workflow

Branching strategy

main          ← stable, always reflects current thesis state
  └── chapter/N-title     ← one branch per chapter draft
  └── sim/feature-name    ← simulation development branches
  └── fix/description     ← corrections and revisions

Commit conventions

[chapter-N] Short description of change
[sim] Short description of simulation update
[lit] Literature / research update
[mem] master_memory.md update
[fix] Correction or revision

Push to GitHub

# Set token (never hardcode in files)
$env:GITHUB_TOKEN = "your_token_here"

# Stage, commit, push
git add -A
git commit -m "[mem] Update master memory — session YYYY-MM-DD"
git push origin main

---

Conventions

• Filenames — snake_case, date-prefixed where chronological (YYYY-MM-DD_topic.md)
• Atomic claims — each entry in extracted_claims.json must carry: id, statement, sources[], confidence, contested (bool)
• Sources — never cite a claim without a peer-reviewed or verifiable anchor; tag uncertain claims explicitly
• Versioning — draft files use _vN suffix; never overwrite past versions during the writing phase
• Token security — never hardcode credentials in versioned files; use $env:GITHUB_TOKEN

---

Key References

• Sentis & Khatib (2005) — Synthesis of whole-body behaviors, IJRR
• Wensing & Orin (2013) — Dynamic humanoid behaviors through task-space control, ICRA
• Koolen et al. (2016) — Momentum-based control framework (Atlas), IJHR
• Carpentier et al. (2019) — Pinocchio C++ library, IROS
• Mastalli et al. (2020) — Crocoddyl: efficient framework for MPC, ICRA

---

Last updated: 2026-04-22 — github.com/David543M/humanoid-wbc-thesis