RDK Board Baseline & Failure Diagnosis

Before answering any RDK problem, confirm the board from on-board facts, not from the user's words. The board determines every downstream answer — package names, TROS path, toolchain, camera node, default IP. Guessing the board (e.g. assuming X5) is the single most common way these answers go wrong.

Sources: official D-Robotics docs rdk_doc / rdk_x_doc / rdk_s_doc (08_FAQ + 01_Quick_start/install_os), the OpenExplorer / 天工开物 (TianGong) toolchains, and reproduced community cases. Facts are carried over with provenance; nothing is invented.

The one move that matters most

Run a read-only identity probe first. Never flash, upgrade, or drive GPIO before you know the board:

cat /sys/class/socinfo/board_id     # X5 / S-series read this (srpi-config, hobot_mipi_cam use it)
cat /sys/class/socinfo/som_name     # X3 docs use som_name; pick one per board
cat /proc/device-tree/model         # literal model string; fallback /etc/board_config.json
cat /etc/version                     # OS/RDK image version (2.1.0+ also has `rdkos_info`)

If the output doesn't map cleanly to X3/X5/Ultra/S100/S100P/S600, ask for more logs — do not assume X5.

Board baseline cheat-sheet

BPU march is the hb_compile --march / hb_mapper arch flag. bayes (Ultra) ≠ bayes-e (X5) and nash-e/nash-m/nash-p (S100/S100P/S600) are all distinct — artifacts never interchange across marches. Board runtime: X3/Ultra hobot_dnn; X5 hbm_runtime (RDK OS 3.5.0+) or pyeasy_dnn; S-series hbm_runtime.

Default login on every RDK image: both sunrise/sunrise (normal) AND root/root (super). Desktop autologin uses sunrise; RDK Studio's SSH channel often uses root. (FAQ "默认登录账户" Q.)

X-series produce .bin; S-series produce .hbm. Cross-architecture artifacts are never interchangeable. Toolchain (hb_mapper for X, hb_compile for S) runs on an x86 Docker host, never on the board.

Correct the misconception first

When a user opens with a wrong premise, clarify in one line before continuing — don't answer the wrong question. Common premises to catch: "X3 .bin runs on X5" (no — different BPU march), "apt install hb_mapper on the board" (no — host Docker only), "RDK Studio = the board" (no — desktop IDE), "RDK is Horizon's" (current brand is D-Robotics), "Ultra = overclocked X5" (no — bayes ≠ bayes-e), "hobot_dnn in venv" (no — system Python only), "apt upgrade 1.x→3.x" (no — reflashing required), "S600 has a 40PIN" (no — 1.8V self-locking connectors), "CAN on S100 = SocketCAN" (no — MCU-domain CANHAL). Full catalog → hardware-notes.md §2.

Failure quick-routing: symptom → entry point

Match the error keyword, then go to failure-hints.md for the exact command + doc (59 entries).

• Camera — SIGABRT / exit code -6 / No image data / VIDIOC_* / timeout → 99% YUYV, switch to MJPEG; default node /dev/video8
• Model / BPU — No such file *.bin/.hbm / OOM / model incompatible / unsupported op → use find for absolute path; cross-march artifacts don't interchange
• TROS / ROS2 — ros2: command not found / NO_PUBKEY / setup.bash missing → source humble (X3–S100P) or jazzy (S600)
• GPIO / I2C / serial / PWM — Permission denied / gpiochip not found → pin not muxed; use gpiofind (not Pi numbers); serial needs dialout group
• CAN — can0 not found → X5 only has SocketCAN; S-series routes through MCU-domain CANHAL (no can0 netdev)
• Power / hang — under-voltage / throttled / kernel panic / System halted → check power supply (X3 5V/3A+, X5 5V/5A, S100 12–20V, S600 12–28V), then temp
• Network / SSH — Connection refused / timeout / Host key verification failed → ip addr + ping; ssh-keygen -R after reflash; fresh S100/S600 → fixed eth1 192.168.127.10
• Device unrecognized — No such device / ENODEV → dmesg | tail -50 && lsusb && ls /dev/tty* /dev/video* /dev/i2c-* /dev/snd/

Diagnostic command safety tiers

Classify before running (full library):

• safe (read-only, run directly): BPU/SoC monitoring, free/df/dmesg/journalctl, cat /sys/..., ip addr. Universal fallbacks: cat /sys/devices/system/bpu/bpu0/ratio (BPU load), hrut_somstatus (temp/voltage/clock).
• moderate (changes state, confirm intent): apt/pip/npm install, systemctl, nmcli, docker run, insmod/modprobe, kernel-header builds.
• dangerous (can destroy the system / needs explicit OK): dd, mkfs, rm -rf /, fdisk/parted, flash erase, modules_install/depmod, writes to /boot · /lib/modules · /etc/fstab, bootloader/initramfs changes. Never run these before the board is confirmed.

Workflow 1 — Confirm board & baseline

Trigger: identify board / 确认板型 / board profile / connected a new device / 不知道是哪块. Precondition: you have a shell (SSH, serial, or RDK Studio).

1. Read board identity [safe] — run bash scripts/board_probe.sh for structured JSON output (board_id, som_name, model, os_version), or use the raw identity probe at the top of this file. Map output to X3/X5/Ultra/S100/S100P/S600. Unknown output → keep collecting logs.
2. Read OS/image version [safe] — cat /etc/version (and rdkos_info on 2.1.0+). Package names, TROS path, and toolchain advice all depend on this.
3. Pick the BPU monitor by board [safe] — X3: hrut_smi; X5/Ultra: hrut_bpuprofile -b 0; S-series: hrut_bpuprofile; universal fallback: cat /sys/devices/system/bpu/bpu0/ratio. (hrut_smi/bputop are NOT on X5.)
4. State the baseline back — name board, SoC/BPU arch, OS version, and what's still unknown before giving any install/deploy/camera/TROS advice. Don't apply the X5 template to an unconfirmed board.

Safety: read-only only; no flashing/upgrade/GPIO output before the board is confirmed.

验证: bash scripts/board_probe.sh outputs JSON with board_id matching the expected board (X3/X5/Ultra/S100/S100P/S600); cat /etc/version confirms OS image version; BPU monitor command (hrut_smi / hrut_bpuprofile / cat /sys/devices/system/bpu/bpu0/ratio) returns a non-zero ratio.

Workflow 2 — S-series xburn flashing (DFU / Fastboot)

Trigger: flash S100/S100P/S600 / xburn / enter download mode / DFU / Fastboot / bricked / blank-board flash. (X3/X5/Ultra use SD card, not xburn.) Full step-by-step: xburn-flashing.md.

1. Host prep — PC USB ↔ board Type-C, high-quality short shielded cable. Linux: apt install android-tools-adb android-tools-fastboot dfu-util. Windows: sunrise5_winusb driver + CH340 serial driver (921600/8/N/1/none). Full host setup → xburn-flashing.md §2.
2. Choose mode — DFU+Fastboot (blank/bricked, must set hardware into DFU) vs Fastboot (normal update, U-Boot boots or fastboot 0).
3. Enter DFU — per-board switch sequence (S100/S100P: SW1+SW2 combo; S600 V0P1: jumper; V0P2: FLASH switch) → xburn-flashing.md §4.
4. Xburn settings — product RDKS100/RDKS600, media S100=emmc / S600=ufs, type secure → xburn-flashing.md §5. Region flash/backup → §6–7.
5. Finish — power off, flip boot switch back (exit DFU), power on. First boot ~45 s config; HDMI shows Ubuntu desktop.

Safety: flashing is dangerous (flash erase). Confirm the board matches the product image (don't flash an S100 image onto an S600); always set boot switches/jumpers with the board powered off.

验证: xburn --chip reads the correct chip model; after reboot, bash scripts/board_probe.sh shows the board type changed to the flashed target; HDMI displays the Ubuntu desktop within ~45 s of first-boot config.

Worked examples

Example 1 — "板子连不上，ssh 一直 timeout,是新到的 S100" A fresh S100/S600 has eth1 fixed at 192.168.127.10 (eth0 is DHCP). Tell them: set the PC NIC to the same subnet (e.g. 192.168.127.100/24), then ssh root@192.168.127.10 or ssh sunrise@192.168.127.10. Don't chase Wi-Fi/router config first.

Example 2 — "跑 ros2 launch 报 /opt/tros/humble/setup.bash: No such file,这是 S600" On S600 the premise is wrong: S600 is Ubuntu 24.04 / ROS2 Jazzy, so TROS lives at /opt/tros/jazzy/, packages are tros-jazzy-*. source /opt/tros/jazzy/setup.bash. Humble paths are for X3/X5/Ultra/S100/S100P only.

Example 3 — "USB 摄像头 launch 一跑就 exit code -6 / SIGABRT" Camera node crash, almost always format. Order: ls /dev/video* + lsusb → v4l2-ctl -d /dev/video0 --list-formats-ext to read real modes → set the launch to MJPEG at a resolution that exactly matches a listed entry (validate 640×480 first). Note the RDK USB-cam default node is /dev/video8. Don't tweak other params before the format matches.

Example 4 — "我要给一块变砖的 S600 重新刷系统" Route to Workflow 2 / xburn-flashing.md. Use DFU+Fastboot (blank/bricked). Enter DFU by board rev: V0P1 short the jumper, V0P2 set FLASH ON — both with PWR KEY OFF first, then ON until the FLS red LED lights. Xburn: product RDKS600, media ufs, type secure. Always power off before flipping switches.

Common pitfalls

Anti-hallucination guardrails

When answering from this skill, follow these rules — never fabricate facts, commands, or file paths:

1. Report only observed data. Quote what scripts/commands actually return, not what you remember. If the probe says board_id: X5, answer for X5 — even if the user insists it's an S100.
2. No fabrication when tools are missing. If a script or reference doesn't exist, say "not found" — don't invent from memory. Route to the appropriate skill or doc instead.
3. Preserve null/false/empty on failure. If a probe returns null or false, report that — don't substitute a plausible value. Empty output is data, not an error to "fix".
4. No substitution off-platform. If off_platform: true, say "probe didn't run on an RDK board" — don't guess what it would have returned. Ask for on-board logs.
5. No hand-editing JSON. Scripts emit structured JSON; never hand-craft output. If the contract says {ok,off_platform,reason,fields}, that's what goes to the user.
6. Acknowledge sandbox limits. If you can't run a command, say so — don't pretend you did. Offer the command for the user to run.
7. Read-only boundary. Never modify the system — no dd, mkfs, rm -rf, apt install, reboot, or GPIO output without explicit user confirmation.

Reference map