Skip to content

HISTORY.md

Latest commit

 

History

History
136 lines (117 loc) · 7.17 KB

HISTORY.md

File metadata and controls

136 lines (117 loc) · 7.17 KB
Raw

Nux — Development History

A running log of project work in chronological order (newest entries at the bottom). Release-facing changes are summarised separately in CHANGELOG.md.

2026-08-02

Project inception

  • Chose the project name Nux (the "war boy" from Mad Max: Fury Road), matching the Devilboy — Born to race body livery.

Architecture decisions

  • Hardware direction: STM32F103C8T6, MOSFET H-bridge drive (target DRV8251A), SG90 servo steering, SX1276 GFSK link at 868 MHz, split 2S traction / 1S logic 18650 power, on/off scale lighting.
  • Software architecture: bare-metal C11 (no HAL), cooperative AcroSched scheduler linked as a frozen pre-built library, shared radio protocol compiled into both firmwares.
  • Monorepo layout: common/ (protocol + scheduler libs + adapter), vehicle/ and transmitter/ firmwares.

Documentation

  • Recorded the initial project requirements in requirements.md.
  • Created the initial documentation set (README, requirements, history, changelog, wiki).

Protocol

  • Drafted the shared radio contract common/protocol/nux_protocol.h (little- endian packed frames, CRC-16/CCITT-FALSE): SNuxCommand_t (10 B) and SNuxTelemetry_t (9 B), light/status flag enums, _Static_assert size guards.

2026-08-03

Architecture decisions

  • Protocol draft resolved: axis resolution stays int16 (±1000); telemetry keeps both battery_mv and battery_pct; the software in-payload CRC-16 is kept (end-to-end, independent of the SX1276 hardware CRC); the frame seq is widened uint8 → uint16; the checksum moves into its own nux_crc module.
  • AcroSched build configuration agreed: cooperative kernel, ACROSCHED_MAX_TASKS = 8, 32-bit tick (uint32_t) at 1 kHz; watchdog hook on (IWDG), idle / low-power __WFI hook on, software timers off (periodic task modes cover turn-signal blink, fail-safe and telemetry cadence). Radio RX uses IPC event flags (RX_DONE set by the DIO0 ISR), not the mailbox. Both firmwares link the same library configuration.

Requirements

  • Clarified REQ-NUX-074 (radio RX via IPC event flags). Added REQ-NUX-076 (AcroSched build configuration), REQ-NUX-077 (hardware watchdog) and REQ-NUX-078 (idle / low-power hook).

Protocol

  • Widened the frame seq to uint16; command frame is now 11 B, telemetry 10 B (_Static_assert size guards updated).
  • Extracted nuxCrc16 (CRC-16/CCITT-FALSE) into common/protocol/nux_crc.{h,c}.
  • Implemented the frame API in common/protocol/nux_protocol.c: nuxCommandFinalize/nuxCommandValid, nuxTelemetryFinalize/nuxTelemetryValid.

Scheduler adapter

  • Vendored the frozen AcroSched 2.1.0 library into common/lib/: public headers in inc/ plus the pre-built archives ac6/acrosched.lib and gcc/libacrosched.a. Confirmed the vendored acrosched_config.h matches the agreed configuration (cooperative, MAX_TASKS = 8, watchdog on, IPC on, timers off, idle hook on).
  • Added the Cortex-M3 port header common/lib/inc/acrosched_port.h (32-bit tick, PRIMASK critical sections, __WFI wait-for-event); it is required because acrosched_config.h includes it and it is not part of the frozen library.
  • Wrote the thin scheduler adapter common/sched/nux_sched.{h,c} (REQ-NUX-072): a self-contained Nux* interface that hides the acro* API, owns the 1 kHz tick counter (nuxSchedTick / nuxSchedNow, bound to the scheduler in nuxSchedInit), forwards task management, and re-exports the IPC event flags as SNuxEventGroup_t / nuxEventSet/nuxEventGet/nuxEventClear. _Static_assert guards keep the mode/status encodings and the tick width in lock-step with the pre-built library ABI.

2026-08-04

Board support package

  • Cleaned the imported BSP skeletons for both firmwares (vehicle/bsp/, transmitter/bsp/) down to a shared, board-agnostic base; the two copies are kept identical for now (per-peripheral split comes later).
  • Retargeted from the STM32VLDISCOVERY template (STM32F100 Value Line, 24 MHz) to STM32F103C8: 72 MHz SYSCLK from the 8 MHz HSE crystal (PLL ×9), Flash 2 wait states + prefetch, APB1 /2 (36 MHz), APB2 72 MHz; debug USART1 on PA9/PA10 re-tuned to BRR = 625 for 115200 baud.
  • Bound SysTick to the scheduler adapter: SysTick_Handler now calls nuxSchedTick(); dropped the BSP-owned sys_tick, the acrosched.h include and the pre-emption tick (cooperative kernel only).
  • Kept the fault handlers (handlers.c) as a naked MSP/PSP trampoline into a common C handler that captures the stacked frame and fault-status registers.

Build system

  • Set up the CMSIS-Toolbox solution nux.csolution.yml with a single target (STM32F103C8), Debug/Release build types and both toolchains (AC6 6.20.1, GCC 15.2.1).
  • Following the intended workflow, projects pull in only ARM::CMSIS:CORE (no Device:Startup); the device header and STM32F10X_MD define come from the DFP via the selected device, and startup/SystemInit are project-owned.
  • Added a shared layer common/nux_common.clayer.yml (CMSIS core, protocol, scheduler adapter and the pre-built AcroSched library — .lib for AC6, .a for GCC) plus the two firmware projects vehicle/transmitter, each with its BSP group and a minimal src/main.c skeleton.
  • Vendored the STM32F103C8 memory map and linker templates into each project's RTE/Device/STM32F103C8/ (regions_*.h, ac6_linker_script.sct.src, gcc_linker_script.ld.src); cbuild wires the linker: node automatically.
  • GCC C runtime resolved with --specs=nano.specs --specs=nosys.specs.
  • Verified: all four contexts build clean on both toolchains; the AcroSched library links in each case (ROM ~2 KB, RAM ~0.8 KB for the skeleton).

Documentation & runtime stubs

  • Silenced the GCC-only newlib link warnings (_close/_lseek/_read/_write "is not implemented and will always fail") by adding minimal no-op syscall stubs bsp/src/syscalls.c to both firmwares, compiled for GCC only (for-compiler: GCC); AC6 retargets through the ARM C Library and never sees the file. Re-verified: GCC 4/4 and AC6 4/4 contexts still build clean.
  • Aligned the README repository-layout tree with the actual structure (nux.csolution.yml, common/nux_common.clayer.yml, lib/{inc,ac6,gcc}, per-project *.cproject.yml, bsp/{inc,src}, src/, RTE/Device/).

2026-08-12

Control strategy decision

  • Fixed the control strategy for the first prototype: the steering servo is driven from a 50 Hz pulse train with direct angle-to-pulse conversion; there is no MCU-side PID loop for the servo.
  • Fixed the traction control law for the v0.1 prototype: no speed sensor is available, so there is no speed PID loop; PWM duty is mapped to throttle and current feedback is used as the active control signal.
  • Chosen regulation model: open-loop throttle mapping with a PI current loop (driver IPROPI or shunt) for load limiting and coarse torque management, with current limiting, battery undervoltage checks and fail-safe active braking.
  • This decision is now reflected in the project requirements so the architecture remains stable while the hardware and firmware are implemented.