Replacement touchscreen wall panels for a 2019 Entegra Aspire 44W with a Firefly Integrations G6A multiplex system.
The coach's factory switch panels are fixed-function membrane switches. This project replaces them with 4.3" capacitive touchscreens that speak the coach's native RV-C protocol directly on the CAN bus — no hub, no gateway, no cloud. Each panel is an independent peer node: it transmits light commands and listens for status frames, so it stays in sync with the factory switches and the Firefly app automatically.
Beyond lights, the panels also show tank levels, a combined battery bank readout for three parallel LiFePO4 packs, and shore power from a Hughes Power Watchdog — the latter two reached over BLE and shared to every panel by ESP-NOW broadcast.
 |
 |
📖 All screen samples → · 🔌 System architecture and installed equipment →
- Status-driven UI. A button lights up only when a
DC_DIMMER_STATUS_3frame from the bus says the load is on — never because you pressed it. Flip a factory switch and this panel updates; press here and the factory panel's LED updates. No master, no polling. - Tap to toggle, press-and-hold to dim. Holding sends RV-C ramp commands while held and a stop on release.
- Multi-instance buttons. One button can drive several RV-C instances (e.g. SIDE CEILING = 30 + 31); it shows ON if any member is on and sends explicit on/off to all members so they can't drift out of sync.
- Runs on the bus at night. Dark theme, idle auto-dim at 120 s and backlight-off at 300 s, and the touch that wakes the screen doesn't trigger the button underneath.
- Data it isn't wired to. A headless ESP32 in the basement bay holds every BLE link in the coach — the three battery packs and the shore-power monitor, all of which sit in that bay — and re-broadcasts their readings over ESP-NOW, so any panel can display them without a connection of its own. No panel runs a BLE stack.
Six nodes: four touchscreen panels, one headless BLE proxy, and one headless valve controller. They reach the coach over RV-C CAN and each other over ESP-NOW. Everything reached over BLE — the battery packs, the shore-power monitor and the solar charge controller, all of which live in the basement bay — is held by the proxy in that bay and broadcast to the panels.
| Node | Board | Talks |
|---|---|---|
mid_coach |
Waveshare ESP32-S3-Touch-LCD-4.3B | RV-C CAN + ESP-NOW bridge |
bedroom_remote |
Waveshare ESP32-S3-Touch-LCD-4.3B | ESP-NOW only |
main_cabinet |
Waveshare ESP32-S3-Touch-LCD-7B (1024×600) | RV-C CAN + ESP-NOW telemetry (listen only) |
| Bluetooth proxy basement | ESP32-D0WD-V3, 4 MB | BLE (3 battery packs + Power Watchdog + Renogy solar) + ESP-NOW broadcast |
valve_node (planned) |
Waveshare ESP32-S3-ETH-8DI-8RO | ESP-NOW unicast — drives the two dump valves |
Panel boards are ESP32-S3-WROOM-1 with 16 MB flash / 8 MB octal PSRAM, GT911
touch, CH422G expander, onboard TJA1051 CAN transceiver and 7–36 V input.
The 4.3B panels run their 800×480 LCD rotated to portrait; main_cabinet's
7B runs landscape at 1024×600, which is what gives its side-nav rail room to
live. (It was an 800×480 non-B 7" until the 7B swap; that board's config
survives as components/board/board_lcd7, unused.)
BOARD is derived from
PANEL in the root CMakeLists.txt and validated by
tools/check_panels.py; there is no -DBOARD= to get wrong.
The two headless nodes are not panels: they carry no display, no LVGL and no
PANEL_INDEX, and each is its own ESP-IDF project pulling shared components
individually.
📖 Full equipment list, protocols and architecture diagram: docs/SYSTEM.md · Dump-valve wiring diagram and build spec: docs/DRAINMASTER-VALVES.md
firefly-touch/
├── main/ # panel app: tasks, state, UI
├── components/
│ ├── rvc_protocol/ # RV-C encode/decode ┐ pure C,
│ ├── jbd_bms/ # battery BMS + bank agg ├ host-testable
│ ├── hughes_watchdog/ # Power Watchdog decode ┘ (host_test/)
│ ├── espnow_link/ # panel-to-panel link + telemetry broadcast
│ ├── board_4_3b/ # Waveshare bring-up: RGB, CH422G, GT911, TWAI
│ └── ui_common/ # theme + shared widgets
├── panels/ # ← per-panel config, selected at build time
│ ├── REGISTRY.md # canonical index → source-address allocation
│ └── TEMPLATE.h # copy this to add a panel
├── proxy/ # basement BLE proxy — SEPARATE project,
│ # classic ESP32 target, no display
│ # (valves/ will join it: dump-valve relay
│ # controller, also headless — see
│ # docs/DRAINMASTER-VALVES.md)
├── sim/ # PC simulator (see the UI without hardware)
├── tools/check_panels.py # enforces unique indices + registry sync
├── docs/ # SYSTEM.md, SCREENS.md, FLASHING.md,
│ # DRAINMASTER-VALVES.md, images
├── .github/workflows/ # CI: lint, host tests, every panel + the proxy
└── CLAUDE.md # architecture, DGN tables, pinout, TODOs
One codebase, many panels. A panel is just a header in panels/ defining
its name, index, and button grid. Adding a third panel is one new header plus
a build flag — no code changes. Copy panels/TEMPLATE.h, claim the next index
from panels/REGISTRY.md, and run
python tools/check_panels.py. Full procedure:
docs/FLASHING.md.
Updates after installation. No OTA — the ESP32-S3's Wi-Fi radio is
WPA2-only and can't associate to a WPA3+PMF network, so updates stay
USB-only by design (single-app partitions.csv, no Wi-Fi/OTA code). See
docs/FLASHING.md.
You need ESP-IDF v5.3+ (developed against v5.5.5). Everything else —
compiler, CMake, Ninja, Python env — comes from ESP-IDF's own installer.
Install both targets: esp32s3 for the panels, esp32 for the proxy.
git clone --branch v5.5.5 --depth 1 --recurse-submodules --shallow-submodules https://github.com/espressif/esp-idf.git C:\esp\esp-idf
C:\esp\esp-idf\install.ps1 esp32s3,esp32# Linux / macOS
git clone --branch v5.5.5 --depth 1 --recurse-submodules --shallow-submodules https://github.com/espressif/esp-idf.git ~/esp/esp-idf
~/esp/esp-idf/install.sh esp32s3,esp32Then in every new terminal, activate the environment before using
idf.py — it is deliberately not on your global PATH:
. C:\esp\esp-idf\export.ps1 # or: . ~/esp/esp-idf/export.shThe ESP-IDF VS Code extension wraps the same toolchain and handles the environment for you.
A host C compiler is needed only for the unit tests and the PC simulator,
not for firmware — on Windows, winget install BrechtSanders.WinLibs.POSIX.UCRT.
Managed components (LVGL 9.5.0, esp_lvgl_port 2.8.0, esp_lcd_touch_gt911
1.2.0) download on first build, pinned by dependencies.lock.
Each panel builds into its own directory. PANEL selects which panel's
buttons and identity are compiled in — see
docs/FLASHING.md for the full per-device procedure.
idf.py -B build_mid_coach -DPANEL=mid_coach -p COM5 flash monitorThe basement proxy is a separate project on a different chip, so it builds on its own:
idf.py -C proxy -B proxy/build set-target esp32
idf.py -C proxy -B proxy/build -p COM4 flash monitor
⚠️ Panels that use ESP-NOW or BLE keep real peer MACs and keys in their per-buildsdkconfig, which is not in git. Pass-D SDKCONFIG=build_<panel>/sdkconfigon every invocation, and never regenerate that file — see docs/FLASHING.md.
Verify which firmware a board is running — it prints its identity at boot and shows the name in the status bar:
I (312) main: firefly-touch panel 'MID COACH' (index 0, RV-C source addr 0x80)
See and click the real UI without flashing anything. It compiles the actual
ui.c and widget code against the same LVGL version the firmware uses, with a
fake RV-C bus that echoes status frames back.
cd sim
.\build.ps1 -Run # mid_coach
.\build.ps1 -Panel bedroom_remote -Run
.\build.ps1 -Panel bedroom_remote -Shot p.bmp -Screen2 # headless captureMouse = touch: click to toggle, click-and-hold to ramp. The window is 480×800, matching the firmware's logical (post-rotation) resolution. Fake tank, battery and shore-power feeds make every state reachable, including a pack dropping offline.
Requires a host C compiler (above); SDL2 downloads into sim/third_party/
on first build. Run one idf.py build first so managed_components/ exists.
See docs/SCREENS.md for what each screen looks like.
Every protocol codec is pure C with no ESP-IDF dependency, so it runs natively on your machine — RV-C frames, the JBD battery protocol and bank aggregation, and the Power Watchdog decode. CI runs all three.
cd components/rvc_protocol/host_test
gcc -Wall -Wextra -Werror -I../include ../rvc_protocol.c test_rvc.c -o test_rvc && ./test_rvcSame pattern in components/jbd_bms/host_test and
components/hughes_watchdog/host_test (both need -lm). Real captured
device frames are committed as regression vectors, so a "simplification"
that breaks a byte offset fails loudly.
Pushing a v*.*.* tag runs the full pipeline and publishes a GitHub Release
with a zip per panel (app, bootloader, partition table, flash_args) for
flashing via docs/FLASHING.md. No OTA and no deploy
target — these are wall panels updated over USB.
RV-C instance numbers came from the factory switch legends and must be
confirmed on the real bus. Sniffer mode logs every frame (raw ID, DGN,
source address, data bytes): idf.py menuconfig → Firefly Touch Panel →
RV-C sniffer mode.
Confirmed on the coach: RV-C lighting control against a real G6, tank levels from the SeeLevel, the ESP-NOW bridge and remote panel, all three battery packs over BLE with the combined bank readout, and the Power Watchdog with values matching the unit's own display.
main_cabinet (the 7" side-nav panel) was confirmed on the coach
2026-08-23: RV-C CAN works on the 7" board's GPIO20/19 pins, the battery
bank and shore power arrive as ESP-NOW broadcasts from the proxy, Light
Master on/off behaves as designed, and the display geometry and colours are
correct. That settled both bring-up unknowns for the non-B board — the RGB
timings and the EXIO5 USB/CAN mux polarity.
The panel was migrated to the 7B (1024×600) in issue #66:
components/board/board_lcd7b with the 7B's own RGB timings (from
Waveshare's official 7B demo), and a larger UI variant gated to that panel
(components/ui_common/include/ui_metrics.h). The 7B is pin-identical to
the non-B for CAN, touch and the EXIO map; what differs is geometry, the
RGB porches, the IO expander (a register-addressed "IO_EXTENSION" at 0x24,
not a CH422G — components/ws_io_expander), an extra EXIO6 = LCD_VDD_EN
panel-supply pin, and the pixel clock (16 MHz on the non-B; the 7B panel is
nominally 30 MHz but this firmware runs it at 21 MHz — the
avoid_tearing + full_refresh buffer scheme it needs can't keep ahead of
a faster scanout). Bench-verified 2026-09-05: boots and renders all four
sections stably at 1024×600. Coach sign-off (touch alignment, CAN to the
G6, the 300 s idle-off stage) is still pending.
Known open items, all tracked as TODO(bench) in code:
- The G6's own LIGHT MASTER command, if it has one. The panel's master is synthesised (sweep the on-lights off; apply a scene for on) because the factory rocker's DGN has never been captured. Worth sniffing the rocker to find out whether a single real frame exists.
- Power Watchdog byte offsets came from public reverse engineering rather than a capture from this unit; the values check out against its display, but the client logs raw packets for confirmation.
- The rest of the instance map, the SECURITY (patio/hitch) button, and the PANEL LIGHTS (PL1) DGN.
CLAUDE.md holds the architecture, DGN tables, pinout and the full TODO list. CHANGELOG.md records what's built, what's verified, and what is still assumption.
MIT — Copyright (c) 2026 Garrett.
RV-C is an open standard published by the RV Industry Association. This project is not affiliated with or endorsed by Firefly Integrations, Entegra Coach, Waveshare, Vatrer, Garnet, or Hughes Autoformers.
No warranty. This firmware transmits on a live vehicle control bus; you are responsible for verifying it against your own coach before connecting it.