Primary-Source Registry / 权威来源¶

This registry anchors the foundations, knowledge graph, and environment layers to primary papers, official documentation, or author-maintained textbooks. It was reviewed on 2026-08-22. External APIs and software versions can change; the repository lock files and recorded experiment environment remain the source of truth for reproduced runs.

01 Python and numerical computing¶

Python Tutorial — Python Software Foundation.
NumPy user guide — NumPy project.

02 Linear algebra¶

Introduction to Applied Linear Algebra — Stephen Boyd and Lieven Vandenberghe.
NumPy linear algebra reference — NumPy project.

03 Deep learning¶

PyTorch: Learn the Basics — PyTorch project.
Deep Learning — Goodfellow, Bengio, and Courville.

04 Transformers¶

Attention Is All You Need — Vaswani et al.
PyTorch Transformer documentation — PyTorch project.

05 Coordinate transforms¶

Modern Robotics, rigid-body motions — Lynch and Park, Chapters 2–3.

06 SO3 and SE3¶

Modern Robotics, Chapter 3 — rotation matrices, exponential coordinates, and homogeneous transforms.
SciPy Rotation reference — SciPy project.

07 Kinematics, Jacobians, and IK¶

Modern Robotics, Chapters 4–6 — forward kinematics, Jacobians, and numerical IK.

08 Control¶

Underactuated Robotics — Russ Tedrake, MIT.
Modern Robotics, Chapter 11 — robot motion and force control.

09 MuJoCo¶

MuJoCo documentation — MuJoCo project.
MuJoCo Menagerie — model collection and per-model licenses.

10 Datasets and training¶

LeRobot documentation — Hugging Face.
RLDS specification and tools — Google Research.

11 Probability and optimization¶

Convex Optimization — Boyd and Vandenberghe.
SciPy optimization reference — SciPy project.

12 Perception and sensors¶

OpenCV camera calibration and 3D reconstruction — OpenCV project.
ROS REP 103: Standard Units of Measure and Coordinate Conventions — ROS community standard.

13 Robot systems and safety¶

ROS 2 documentation — Open Robotics.
ROS 2 Quality of Service settings — Open Robotics.

These software references do not certify a physical robot. Applicable laws, standards, robot-manufacturer limits, site rules, and a qualified safety review remain mandatory for hardware use.

14 Evaluation and reproducibility¶

ACM Artifact Review and Badging — Association for Computing Machinery.
DoF benchmark source of truth — machine-readable repository evidence.

15 Multimodal perception and state estimation¶

OpenCV Camera Calibration — intrinsics, distortion, extrinsics, and undistortion.
ROS 2 message_filters — timestamp-based exact and approximate synchronization.
robot_localization — EKF/UKF sensor fusion and state-estimation interfaces.

Nav2 Concepts — behavior trees, planners, controllers, smoothers, routes, and recovery servers.
ROS REP 105: Coordinate Frames for Mobile Platforms — map, odom, and base_link frame semantics.
Isaac Lab environments — maintained navigation and legged-locomotion environment registry.

17 Robot development environment¶

ROS 2 Jazzy Ubuntu packages and ROS 2 Humble Ubuntu packages — official ROS binary installation paths.
Installing Gazebo with ROS and ROS 2 integration — default pairings and ros_gz boundaries.
MuJoCo Python and MuJoCo visualization — maintained bindings, viewer, and rendering contexts.
Isaac Lab local installation — current supported host, Python, Isaac Sim, driver, and installation routes.
Genesis World installation and minimal scene — current package and execution sequence.
Python venv, pip repeatable installs, and PyTorch local selector — isolation and framework-wheel selection.
CUDA Linux installation, CUDA on WSL, and Microsoft WSL installation — driver, toolkit, and WSL boundaries.

18 Rigid-body dynamics and contact¶

Modern Robotics, dynamics chapters — rigid-body dynamics, inverse dynamics, and trajectory generation.
MuJoCo computation — equations of motion, actuation, constraints, contact frames, friction, and solver semantics.
MuJoCo modeling — model parameters, contact parameter combination, and modeling guidance.

19 State estimation and motion planning¶

Underactuated Robotics: State Estimation — observers, Kalman filtering, recursive Bayesian filtering, and smoothing roadmap.
Underactuated Robotics: Sampling-based Motion Planning — graph search, PRM, RRT, feasibility, and dynamic constraints.
Nav2 concepts — maintained planner, controller, behavior-tree, recovery, and navigation-system boundaries.

20 Imitation learning, VLA, and cross-embodiment¶

ACT: Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware — original action-chunking Transformer paper and evaluation scope.
Diffusion Policy project and paper — action diffusion, receding-horizon execution, benchmarks, code, and data from the authors.
Open X-Embodiment / RT-X — original dataset-mixture, action convention, model, and cross-robot evaluation report.
OpenVLA — original project page with paper, code, checkpoints, architecture, adaptation protocol, and reported evaluations.
Octo — original open generalist policy paper and reported cross-embodiment scope.
π0 — original flow-based vision-language-action model report.
OpenVLA-OFT — original report on continuous action representation, parallel decoding, action chunking, and fine-tuning.
SmolVLA and official LeRobot guide — original model report and maintained usage boundary.
LeRobot documentation — maintained policy, dataset, simulation, evaluation, and hardware interfaces; pin a release or commit for reproduction.

21 Reinforcement learning and post-training¶

Reinforcement Learning: An Introduction — Sutton and Barto's author-maintained textbook.
Proximal Policy Optimization Algorithms — original PPO objective and experiments.

22 World models and predictive control¶

Learning Latent Dynamics for Planning from Pixels — original PlaNet latent-dynamics planning paper.
Mastering Diverse Domains through World Models — original DreamerV3 paper and reported evaluation scope.
TD-MPC2: Scalable, Robust World Models for Continuous Control — original TD-MPC2 paper on learned models and planning.
V-JEPA 2 — original latent video representation and action-conditioned planning report; the action-conditioned planner is not labeled a WAM in this repository.
WorldVLA — original autoregressive action-world model preprint unifying image and action understanding/generation.
DreamZero — 2026 preprint on a video-diffusion-based World Action Model jointly predicting future video and actions.
Action Images — emerging 2026 preprint on action-grounded video generation; treat publication status and conclusions as preprint evidence.

23 Manipulation, dexterity, and locomotion systems¶

Modern Robotics — kinematics, dynamics, motion planning, control, and manipulation foundations.
Underactuated Robotics — dynamics, trajectory optimization, state estimation, control, and locomotion notes.
MuJoCo contact computation — point-contact frames, friction dimensions, margins, gaps, and constraint-force semantics used by the simulation guides.

24 Sim-to-Real and deployment evidence¶

Sim-to-Real Transfer of Robotic Control with Dynamics Randomization — original dynamics-randomization study and its stated robot-task scope.
Isaac Lab manager-based environment tutorial — maintained EventManager interface used to schedule startup, reset, interval, and other environment events, including randomization; verify the installed Isaac Lab version before use.
DoF validation policy — repository-local evidence ladder separating synthetic input, simulation, benchmark, HIL/shadow, and hardware claims.

Citation policy¶

Prefer the original paper or official project documentation over secondary summaries.
Record the accessed version/date when an external interface can drift.
Treat a URL check as availability evidence only; it does not prove that every local interpretation is semantically correct.
Open a correction issue when a source and repository statement disagree.