Foundations Layer · English Overview¶
This page gives English-speaking learners a complete route through the 14 prerequisite lessons. The linked lessons currently use Chinese-first explanations, equations, runnable code, and exercises; this overview makes the learning contract, prerequisites, artifact, and exit criterion explicit in English.
Route¶
Python → Linear Algebra → Deep Learning → Transformers
└──→ Frames → SO(3)/SE(3) → FK/Jacobian/IK → Control → MuJoCo
└──→ Data → Probability/Optimization
→ Perception → Systems/Safety → Evaluation
| # | Lesson | What you must learn | Runnable evidence | Exit criterion |
|---|---|---|---|---|
| 01 | Python for robotics | Arrays, shapes, functions, data flow, debugging | NumPy robotics examples | Explain and verify every tensor shape |
| 02 | Linear algebra | Vectors, matrices, norms, projections, SVD | Geometry examples | Derive and test a least-squares solution |
| 03 | Deep learning | Forward pass, loss, autograd, optimization, overfitting | Small PyTorch MLP | Train, validate, and diagnose a curve |
| 04 | Transformers | Tokens, attention, masks, sequence modeling | Minimal attention example | Trace shapes through one attention block |
| 05 | Coordinate transforms | Frames, homogeneous transforms, composition order | Transform-chain example | State every frame and verify round trips |
| 06 | SO(3) and SE(3) | Rotation representations, Lie groups, interpolation | 3D finger-chain example | Convert representations without singularity mistakes |
| 07 | FK, Jacobian, and IK | Forward kinematics, Jacobians, numerical IK, constraints | FK/IK and retargeting examples | Reach targets while respecting limits |
| 08 | Control basics | Feedback, PID, trajectories, stability, saturation | Safety-filter and retargeting examples | Tune a bounded loop and explain failure modes |
| 09 | MuJoCo basics | Models, state, stepping, contacts, actuators | MuJoCo and retargeting examples | Run deterministically and inspect contacts |
| 10 | Dataset and training | Episode schema, synchronization, splits, normalization | Canonical dataset and LeRobot adapters | Detect leakage and reproduce a split |
| 11 | Probability and optimization | Random variables, estimation, gradients, constrained objectives | RSSM and optimization examples | Define objective, constraints, and uncertainty |
| 12 | Perception and sensors | Camera geometry, timing, calibration, multimodal observations | Observation-schema example | Pass calibration and synchronization gates |
| 13 | Robot systems and safety | Middleware, rates, watchdogs, state machines, logging | Model interface and safety filter | Demonstrate bounded failure behavior |
| 14 | Evaluation and reproducibility | Protocols, seeds, confidence, ablations, provenance | Benchmark runner | Reproduce a result and state its evidence level |
How to use the lessons¶
- Read the English contract above and the linked lesson.
- Run the referenced project file rather than copying an isolated snippet.
- Complete the “检查理解” questions.
- Record the command, environment, seed, artifact, metric, and failure case.
- Continue only when the lesson exit criterion is satisfied.
Authoritative references for every lesson are maintained in the primary-source registry. Repository claims and experiments follow the validation policy.