Tutorials — Learning Through 48 Examples¶

XM10's features are covered through 48 progressively structured examples. Each folder contains source code and a README, and every README follows the same format: Goal → Prerequisites → Key Code → Experiments → Next Steps + Common Mistakes. The introductory examples (Ex.09 and below) are designed to complete in under 30 minutes. Control and advanced examples (Ex.11 and above) may take anywhere from 45 minutes to several weeks, depending on difficulty.

The full example catalog is at examples/README.md. If you get stuck, check the "Common Mistakes" section at the bottom of each README first. Claude Code users can simply say "Ex.XX isn't working" and the example-helper skill will respond with the relevant common mistakes and troubleshooting documentation.

Recommended Learning Paths¶

Choose the path that matches your current level.

Path	Target Audience	Sequence	Estimated Time
Beginner	New to embedded systems — just get it working	Ex.00 → 01 → 04 → 07 → 09 → 11	About 3 hours
Intermediate	Familiar with embedded basics — communication, real-time control	Ex.02 → 05b → 08 → 09 → 12 → 14 → 18	About 1 week (1–2 hours/day self-study)
Advanced	Control algorithms + hands-on AI design	Ex.03 → 09 → 15 → 16 → 17 → 19 → 20+	About one semester (1–2 weeks per example from Ex.20+)
Physical AI Applications	Rehabilitation, learning-based control, etc.	Ex.21 → 31 → 32 → 33 → 36	Self-directed (difficulty: ⭐⭐⭐)

Estimated times are for typical users. The "one semester" estimate for the Advanced path assumes spending 1–2 weeks per example for in-depth study. A quick run-through is possible in a few days, but following the variant experiments and cited papers takes significantly longer.

Sample 16-Week Course Schedule¶

A suggested semester schedule for university courses. Adjust freely based on student background and board delivery timing.

Week	Topic	Examples	Assignment / Deliverable
1	Environment setup + first build	Ex.00	Video demo of board LED turning on
2	Button + LED + state machine	Ex.01 → 02 → 03	Add 4 custom LED modes of your own
3	External GPIO + ADC	Ex.04 → 05 → 05a	Control 4 LEDs using an external switch
4	Multi-channel ADC + safety switch	Ex.05b → 05c → 06	Demo simultaneous measurement of 8 FSR channels
5	USB serial communication	Ex.07 → 08	Real-time sensor monitoring via PC terminal
6	Binary streaming + xm10 tool	Ex.09	Capture 4-channel graph + analysis report
7	USB-CDC data capture + storage	Ex.09 + xm10 tool	Stream 10 minutes of data → save as .xmlog → export CSV for analysis
8	Midterm / Mini Project 1	(open)	"My board, my data" mini project presentation
9	KIT H10 exoskeleton basic modes	Ex.11 → 12 → 13	Comparison video of 3 operating modes
10	PD real-time control	Ex.14	Tune your own PD gains + compare step responses
11	Debugging + memory patterns	Ex.18 → 19	Health Dashboard result analysis
12	Gait intent recognition + Tiny AI	Ex.16 → 17	7-phase FSM demo + intent signal graph
13	Impedance + transparent mode	Ex.20 → 21	Compare virtual spring stiffness variations
14	Gait phase adaptive torque	Ex.23 → 25	Measure gait phase estimation accuracy
15	Learning / adaptive control	Ex.26 or 27	ILC convergence curve or MRAC adaptive trajectory
16	Final Project	(open application)	Team project presentation + GitHub submission

Teaching tips: - The first 2 weeks are where students struggle most with environment setup. Schedule dedicated time for TA or instructor-led check-ins. - The Week 8 mini project is effective for maintaining motivation. Run it as an open-ended prompt like "build anything using what you've learned from Ex.05–09." - For Week 16, teams of 3–4 are recommended. Teams that dive deep into the Physical AI application track (Ex.21, 31, 32, 33) at the end tend to produce the strongest final projects.

Examples not in this schedule (Ex.05d, 10, 15, 22, 24, 28–36, 38–44) are naturally left for self-directed study or a follow-on course in the next semester.

Part 0 — Board Verification¶

Example	Title	Difficulty	What You Learn
00	Board smoke test	⭐	State machine + LED + USB serial — no external hardware required

Part 1 — XM10 Basic I/O Control¶

From fundamental I/O to state-based programming (FSM) — the core of embedded control.

Button & LED (Ex.01 ~ 03)¶

Example	Title	Difficulty	What You Learn
01	Button + LED basics	⭐	Poll button state and control LEDs
02	Events + special effects	⭐⭐	Detect click events, single-shot LED blink
03	State machine (FSM)	⭐⭐	Long-press mode switching, per-state behavior separation

External I/O — GPIO & ADC (Ex.04 ~ 06)¶

Example	Title	Difficulty	What You Learn
04	External digital control	⭐⭐	External switch/LED on expansion port
05	Analog sensor introduction	⭐⭐	Voltage measurement on fixed ADC pins
05a	DIO→ADC dynamic switching	⭐⭐	Read FSR by switching a DIO pin to ADC mode
05b	FSR 8-channel batch read	⭐⭐	8 channels + resolution configuration
05c	Mixed ADC 12-channel	⭐⭐⭐	Fixed 4 + DIO→ADC 8
05d	DIO/ADC hybrid	⭐⭐⭐	Digital and analog simultaneously
06	Safety state machine	⭐⭐	Limit-switch-based FSM

Recommended ADC series order: Ex.05 → 05a → 05b → 05c → 05d (progressive difficulty)

Part 2 — USB Communication¶

Exchange real-time messages with a PC and stream data. Essential for debugging and data collection. On Rev 2.0 you can also send structs with field names and units, or save data to a USB memory stick.

Note: The USB serial (CDC) port can only be opened by one program at a time. For Ex.07–09 and Ex.44, make sure only one of a serial terminal, PhAI Studio or the xm10 tool is open. Running them simultaneously will cause a conflict.

USB Serial Communication — Ex.07 ~ 09¶

Example	Title	Difficulty	What You Learn
07	USB serial basics	⭐⭐	Send text messages to a PC terminal
08	Sensor data monitoring	⭐⭐	Real-time `sprintf` output of sensor data
09	High-speed binary streaming	⭐⭐⭐	PhAI V2.2 protocol, 1 kHz transmission (view the user channel with the `xm10` tool)

To save and analyze a captured stream on the PC, use PhAI Studio recording or the xm10 tool in the repo — python pc-data-tool/xm10.py recv shows a live graph while the raw stream is saved as .xmlog, then export pulls out CSV (guide). PhAI Studio is still under development, so for now use the xm10 tool to view and save your own (custom) data structs, such as the ones you send with XM_SendUsbDataWithId. (On Rev 2.0 you can also save directly to a USB memory stick — see Save Data to a USB Memory Stick. On Rev 1.1, on-board file storage was removed in v2.5.0 and is not supported.)

Struct Logging — Ex.44¶

Example	Title	Difficulty	What You Learn
44	Send a struct to the PC with field names, units and types 🛑 Rev 2.0 only	⭐⭐⭐	Register a struct once and its field names, units and types travel with the values — integers, `bool`, arrays, scale factors, and a different rate per struct (view it with the `xm10` tool)

Saving to a USB Memory Stick — Ex.10 · 34¶

There is only one USB-C port. While a USB memory stick is plugged in, PhAI Studio, the xm10 tool and firmware upload over USB are unavailable. Unplug the stick and connect the PC when the save is finished. (not yet confirmed on the board)

Example	Title	Difficulty	What You Learn
10	Save values to a USB memory stick 🛑 Rev 2.0 only	⭐⭐⭐	Start and stop saving with a button, save one struct at 100 Hz, insert markers, check missing samples in the statistics
34	Save H10 gait data to a USB memory stick at 1 kHz 🛑 Rev 2.0 only	⭐⭐⭐	Save the H10 gait data (96 B) every 1 ms — body data, checking missing samples

Part 3 — KIT H10 Exoskeleton Basic Modes¶

The three fundamental operating modes of the KIT H10 exoskeleton robot.

Example	Title	Difficulty	What You Learn
11	Passive Mode	⭐⭐⭐	Automatic reciprocal motion using predefined movement commands
12	Active Assist Mode	⭐⭐⭐	Detect user intent and apply assistive torque
13	Resistive Mode	⭐⭐	Built-in H10 mode — resistance sensation like walking through water

Part 4 — Control Algorithm Fundamentals¶

Your first hands-on control algorithms.

Example	Title	Difficulty	What You Learn
14	PD real-time torque control	⭐⭐⭐	PD formula, discrete differentiation, torque saturation
15	Inverted pendulum gait assist	⭐⭐⭐	Gravity compensation MgL·sin(θ) + Lyapunov stability
16	Tiny AI sensor fusion	⭐⭐⭐	On-board 3-layer neural network inference
17	Gait intent recognition (FSM)	⭐⭐⭐	7-phase gait state machine + per-phase torque
18	System debug monitor	⭐⭐	Loop execution time measurement, status dashboard
19	Memory-aware design	⭐⭐⭐	Ring buffer, pool allocator, malloc-free implementation

Part 5 — Advanced Control Algorithms¶

Interaction dynamics, transparent mode, and gait-phase-based control — the canonical techniques of exoskeleton control.

Example	Title	Difficulty	What You Learn
20	Hogan impedance control	⭐⭐⭐	Virtual spring-damper interaction (Hogan 1985)
21	Gravity compensation (transparent mode)	⭐⭐⭐	Mgl·sin(θ) + friction compensation + gradual activation
22	CPG adaptive oscillator	⭐⭐⭐	Automatic gait rhythm synchronization (Ronsse 2011)
23	Gait phase adaptive torque	⭐⭐⭐	4-segment sinusoidal torque profile (Quinlivan 2017)
24	Virtual constraint (HZD)	⭐⭐⭐	5th-order Bézier trajectory (Westervelt 2003)
25	Stance-phase variable stiffness	⭐⭐⭐	Smooth stance/swing transition (Collins 2015)

Part 6 — Learning + Adaptive Control¶

Control that learns from each gait cycle or automatically adapts to changes in the user.

Example	Title	Difficulty	What You Learn
26	Iterative Learning Control (ILC)	⭐⭐⭐	Learn torque profile from each cycle (Emken 2007)
27	Model Reference Adaptive Control	⭐⭐⭐	Online gain adaptation via MIT Rule (Slotine 1991)
28	Admittance control	⭐⭐⭐	Force input → position output (the dual of impedance, Keemink 2018)
29	Left/right coordination control	⭐⭐⭐	Anti-phase symmetry + affected-side reinforcement (rehabilitation)
30	FF + FB hybrid control	⭐⭐⭐	Model-based feedforward compensation + PD feedback

Part 7 — Physical AI Applications¶

The key stages of Physical AI — transparency → intent detection → learning → autonomous replay.

Example	Title	Difficulty	What You Learn
31	Disturbance Observer (DOB)	⭐⭐⭐	Estimate remaining disturbances for true transparent mode
32	Gait intent via foot contact	⭐⭐⭐	Gait phase estimation from heel strike events
33	Kinesthetic teaching + replay	⭐⭐⭐	Human demonstrates by hand → board replays the motion
35	Multi-layer transparent control	⭐⭐⭐	Real-time switching between transparent / wall / left-right coupling modes
36	On-device kinesthetic learning 🛑 Rev 2.0 only	⭐⭐⭐	Train a small neural network on-board → replay with LQR — Internal Flash UserNV API is supported on Rev 2.0 only
38	Periodic background task	⭐⭐	Offload low-frequency auxiliary work, minimize control loop jitter
39	Task lifecycle management	⭐⭐⭐	Create, suspend, and terminate tasks; RTOS task state machine pattern
40	EMG proportional assist 🛑 Rev 2.0 only	⭐⭐⭐⭐	External 4-channel ADC EMG → envelope → proportional torque, button calibration + 0xF0 streaming to the xm10 tool (EMG competition foundation)
41	IMU Hub attitude dashboard 🛑 Rev 2.0 only	⭐⭐⭐	Up to 6 IMUs' quaternion → Euler (roll/pitch/yaw) conversion, auto connection detection + 0xF0 18-channel (50Hz) streaming to the xm10 tool (CAN-FD sensor hub)
42	EMG Hub biofeedback 🛑 Rev 2.0 only	⭐⭐⭐	Receive hub-processed muscle activation (envelope/MVC%), button calibration + LED feedback and a 0xF0 4-channel (50Hz) live view in the xm10 tool (CAN-FD sensor hub, no motor drive)
43	Two boards talking over serial 🛑 Rev 2.0 only	⭐⭐	Connect two XM10s directly over the External UART (PD5/PD6) — copy only inside the callback, find frame boundaries with a byte state machine

The IMU Hub and EMG Hub are modules under internal development. To use them, please contact us at https://huphailab.com/contact.

Trying an Example¶

Open the .c file for the example you want from the examples/ folder.
Copy its entire contents into XM_Apps/Control_Task/control_task.c.
Build in STM32CubeIDE and upload to the XM10.
Follow the example's README to verify the behavior.

If this is your first build or flash: First Build & Run · For AI-guided setup: Getting Started with Claude Code