← Back to Robotics

Learning Gaps Roadmap

What’s missing, why it matters, and in what order to build it

The Gap Analysis

Existing tracks cover hardware/firmware foundations (electronics → protocols → Zephyr) and advanced C++. Missing tracks are all on the software/algorithm side — which is where the daily RCA work actually happens.

EXISTING                            MISSING
────────────────────────────────    ────────────────────────────────────────
electronics/    (hardware theory)   navigation-estimator/  ← HIGHEST PRIORITY
zephyr/         (firmware)          ros2-handson/           ← HIGH
cpp-advanced/   (C++ depth)         python-scripting/             ← MEDIUM
                                    zephyr/deep-dive/       ← MEDIUM (stub only)
                                    linux-rt/               ← LOW (cpp-advanced/03 partial)

Track 1: Navigation & State Estimation

Priority: HIGHEST Why now: You investigate ESTIMATOR_STATE_INVALID, slip detection, line-sensor covariance blow-up, and EKF divergence tickets daily. You know the robot code deeply (repo memory) but lack the mathematical theory to immediately see why a covariance grows / explodes / converges.

Goal: Be able to look at any estimator log and reason from first principles about whether the observed covariance trajectory is expected or anomalous.

Timeline: 5 weeks · 3–4 hrs/week · no hardware needed

Status: 🔴 Not started Folder: learn/navigation-estimator/

Week Topic Practical Relevance
1 Dead-reckoning & wheel odometry math predict() noise model
2 EKF theory: prediction step, covariance propagation Covariance INF = slipped/collision
3 EKF measurement update: Mahalanobis, innovation gating Line-Sensor update in update()
4 IMU fusion, gyro bias, line-sensor as line-constraint measurement Theta update + gyro correction
5 Diagnosing estimator failures from logs/bags Apply to real tickets

Files to create:

navigation-estimator/
├── 00-learning-plan.md        ← dependency graph, study order, checkpoints
├── 01-dead-reckoning.md       ← odometry math, unicycle model, arc integration
├── 02-kalman-filter.md        ← KF → EKF derivation, predict/update equations
├── 03-measurement-models.md   ← line sensors as constraints, Mahalanobis gating
├── 04-imu-fusion.md           ← gyro integration, bias estimation, line-sensor theta
├── 05-failure-modes.md        ← covariance blow-up patterns, slip vs collision vs delocalize
└── exercises/
    ├── 01-odometry-math.md
    ├── 02-kalman-1d.md        ← build a 1D Kalman filter by hand
    ├── 03-ekf-unicycle.md     ← implement EKF for a unicycle robot in Python
    ├── 04-log-diagnosis.md    ← given a covariance trajectory, identify the failure
    └── 05-robot-specific.md     ← navigation estimator parameter tuning exercises

Track 2: ROS2 Hands-On Exercises

Priority: HIGH Why: zephyr/study-notes/05-jetson-ros2.md (1524 lines) is theory-only. You work with ROS2 topics, nodes, tf2, nav2 every day but haven’t built a systematic exercise set.

Goal: Be able to write, debug, and tune any ROS2 node from scratch. Understand DDS/QoS well enough to explain frame drop bugs.

Timeline: 3 weeks · 3–4 hrs/week

Status: 🔴 Not started Folder: learn/ros2-handson/

Week Topic Practical Relevance
1 Nodes, topics, services, actions, lifecycle AMR node architecture
2 tf2 transforms, time sync, QoS profiles TF lookup failures in nav logs
3 Nav2 BT architecture, costmaps, planner/controller RCS navigation stack

Files to create:

ros2-handson/
├── 00-learning-plan.md
├── 01-nodes-topics-actions.md
├── 02-tf2-time-qos.md
├── 03-nav2-architecture.md
└── exercises/
    ├── 01-pub-sub-timer.md
    ├── 02-tf2-broadcaster.md
    ├── 03-action-server.md
    └── 04-nav2-custom-plugin.md

Track 3: Python / Robot Scripting

Priority: MEDIUM Why: You write Python scripts daily (RCA analysis, KB tools, log parsing) but there’s no structured skill-building. Gaps: async patterns, dataclass typing, pandas/polars for log analysis, test coverage for scripts.

Goal: Write analysis scripts that are testable, typed, and maintainable — not just “it works on my machine.”

Timeline: 3 weeks · 2–3 hrs/week

Status: 🔴 Not started Folder: learn/python-scripting/

Week Topic Practical Relevance
1 Type hints, dataclasses, Pydantic for config validation session_tracker.py, knowledge_search.py
2 pytest, fixtures, mocking for CLI scripts Script tests in tests/
3 pandas/polars for time-series log analysis Bag CSV analysis patterns

Files to create:

python-scripting/
├── 00-learning-plan.md
├── 01-types-dataclasses.md
├── 02-testing-cli-scripts.md
├── 03-timeseries-analysis.md
└── exercises/
    ├── 01-type-annotations.py
    ├── 02-pytest-fixtures.py
    └── 03-log-analysis-polars.py

Track 4: Zephyr Deep-Dive

Priority: MEDIUM (blocked by hardware) Why: zephyr/deep-dive/ exists but is empty. Meant for hardware-stage deep dives once the STM32 + logic analyzer are in hand.

Timeline: Add content incrementally as hardware is acquired (see zephyr/00-mastery-plan.md hardware list)

Status: 🟡 Stub created, blocked on hardware Folder: learn/zephyr/deep-dive/

Files to create (when hardware arrives):

zephyr/deep-dive/
├── 01-first-build-flash-debug.md   ← blinky to shell, GDB, Segger RTT
├── 02-spi-slave-first-frame.md     ← wire up logic analyzer, capture first byte
├── 03-dma-cache-gotchas.md         ← reproduce the D-cache coherency bug on real HW
├── 04-imu-i2c-reads.md             ← ICM-42688 register read via Zephyr I2C API
└── 05-100hz-spi-bridge.md          ← the full capstone

Track 5: Linux RT Internals

Priority: LOW Why: cpp-advanced/03-rt-linux-programming partially covers this (SCHED_FIFO, mutex, priority inversion). Not a separate track — extend cpp-advanced/03 instead.

Status: 🟢 Partially covered in cpp-advanced/03-rt-linux-programming/ Action: After completing cpp-advanced weeks 1–5, add: - perf and ftrace for latency profiling - /proc/latency_stats interpretation - PREEMPT_RT patch model

PARALLEL TRACKS (can interleave):

NOW
│
├── navigation-estimator/   ← start immediately, highest RCA value
│   Week 1: Dead-reckoning math
│   Week 2: EKF theory
│   Week 3: Measurement models
│   Week 4: IMU fusion
│   Week 5: Failure mode diagnosis
│
├── cpp-advanced/ (continue from wherever you are)
│
IN 6–8 WEEKS
│
├── ros2-handson/           ← after navigation-estimator (builds on the theory)
├── python-scripting/             ← can start in parallel with navigation-estimator
│
WHEN HARDWARE ARRIVES
│
└── zephyr/deep-dive/       ← hardware-gated, builds on existing zephyr/ theory

Quick-Start: Next Session

Start here → navigation-estimator/00-learning-plan.md

Day 1 goal (2 hrs): Read 01-dead-reckoning.md and be able to derive (x', y', θ') from two encoder counts for a differential-drive robot.

This directly ties to predict() in navigation estimator — you’ll recognize every variable.