Skip to content

ROADMAP.md

Latest commit

 

History

History
233 lines (178 loc) · 10.7 KB

ROADMAP.md

File metadata and controls

233 lines (178 loc) · 10.7 KB
Raw

Micro-BLSS Development Roadmap

Current Version: 0.2.1 · License: MIT · Last Updated: 2026-05-04

This roadmap outlines the development plan for Micro-BLSS, organized by priority order. Each milestone builds upon the previous one to systematically evolve the project from a functional simulation MVP to a hardware-integrated, AI-driven closed-loop life support digital twin.

Milestone Overview

  v0.1.0 ✅   Initial MVP — MEC plant model, stability monitoring, V&V suite
       │
  ─────┼──────────────── Phase 1: Mathematical Fidelity ────────────────
       │
  1A ✅│   V-HAB 5×5 CQY/T_A polynomial matrices + age-dependent CUE_24
  1B ✅│   Multi-crop parameter library (9 MELiSSA crops)
  1C ✅│   Crew activity scheduling (sleep/nominal/active daily cycle)
  1D ✅│   Oscillation detection (FFT) + Phase-plane analysis
       │
  v0.2.0 ✅   Mathematical Fidelity Release (207 tests)
       │
  1E ✅│   NASA JPL "Power of Ten" safety-critical refactoring
  1F ✅│   Property-based testing (Hypothesis) + Robustness suite
       │
  v0.2.1 ✅   Safety & Robustness Release (221 tests, 91.5% coverage)
       │
  ─────┼──────────────── Phase 2: Hardware-In-The-Loop ─────────────────
       │
  2A   │   MQTT sensor backend + bridge module
  2B   │   ESP32 reference hardware + Live Mode dashboard
       │
  v0.3.0     Hardware-In-The-Loop Release
       │
  ─────┼──────────────── Phase 3: AI-Driven Control ────────────────────
       │
  3A   │   Model Predictive Control (MPC) optimizer
  3B   │   RL agent + pre-built mission scenario library
       │
  v0.4.0     Autonomous Control Release

Phase 1 — Mathematical Fidelity & Multi-Crop (v0.2.0) ✅

1A · V-HAB CQY/T_A Polynomial Matrices + Age-Dependent CUE ✅

Status: ✅ Complete — merged into main.

What was implemented:

  • V-HAB empirical 5×5 polynomial coefficient matrices for CQY_Max and T_A computation, replacing simplified Michaelis-Menten approximation
  • Age-dependent CUE_24 interpolation with linear senescence decay (T_Q → T_M) for both legumes and non-legumes
  • Photoperiod-aware light/dark cycling based on crop-specific fPhotoperiod
  • Biomass growth hard capping at 20% per step
  • Golden reference CSV regenerated with 40 test vectors (up from 21)

1B · Multi-Crop Parameter Library ✅

Status: ✅ Complete — src/modules/crops/ package.

What was implemented:

  • CropParameters dataclass with compute_cqy_max() and compute_t_a_seconds() polynomial evaluators
  • 9 MELiSSA-reference crops: Lettuce, Wheat, Soybean, Rice, Tomato, White Potato, Sweet Potato, Dry Bean, Peanut
  • CROP_REGISTRY dictionary and get_crop() lookup function
  • All coefficient matrices extracted directly from V-HAB MATLAB source
  • Parametrized tests verifying all 9 crops produce positive O₂ at nominal conditions

1C · Crew Activity Scheduling ✅

Status: ✅ Complete — ActivitySchedule dataclass.

What was implemented:

  • ActivitySchedule dataclass with configurable daily cycle (8h sleep, 14h nominal, 2h active)
  • Per-crew-member phase offsets for staggered multi-crew scenarios
  • CrewCompartment automatically cycles activity levels based on simulation clock
  • Backward-compatible static mode (no schedules) preserved
  • Simulation constructor auto-generates staggered schedules for num_crew > 1

1D · Oscillation Detection + Phase-Plane Analysis ✅

Status: ✅ Complete — StabilityMonitor rewrite.

What was implemented:

  • FFT-based oscillation detection on a rolling window of Cᵢ history
  • O₂/CO₂ phase-plane trajectory analysis (converging, diverging, limit_cycle, stable)
  • Dominant period and energy concentration metrics
  • get_phase_plane_data() for dashboard plotting
  • Tests for synthetic oscillatory and stable signals

Phase 1.5 — Safety & Robustness (v0.2.1) ✅

Status: ✅ Complete.

What was implemented:

  • NASA JPL "Power of Ten" Compliance:
    • Refactored all modules to use immutable NamedTuple state objects (Rule 3).
    • Achieved high assertion density for physical invariants (Rule 5).
    • Implemented loop safety bounds and memory history limits (Rule 2).
  • Property-Based Testing: Integrated Hypothesis for BufferReservoir verification.
  • Boundary Safety: Replaced assertions with ValueError for external boundary checks to ensure production persistence.
  • Coverage Boost: Reached 91.5% total test coverage.

Phase 2 — Hardware-In-The-Loop (v0.3.0)

2A · MQTT Sensor Backend + Bridge

Priority: 🟡 High — enables the physical-digital bridge.

Changes:

  • Implement dual-mode sensor backend in src/core/sensors.py:
    • MockSensorBackend (existing Gaussian noise)
    • MQTTSensorBackend (subscribes to MQTT topics via paho-mqtt)
  • Create src/io/mqtt_bridge.py with:
    • MQTT client connecting to configurable broker (default: localhost:1883)
    • Subscribe topic schema: habitat/sensor/{o2,co2,humidity,temperature}
    • Publish topic: habitat/twin/state (JSON with full simulation state)
    • Automatic reconnect and QoS 1 for reliability
  • Add --real-sensors CLI flag to simulation.py for MQTT mode
  • Add --broker-host and --broker-port CLI arguments
  • Write integration tests with a mock MQTT broker

Verification: Simulation runs in MQTT mode, receives mock sensor data from a test publisher, and publishes twin state back.

2B · ESP32 Reference Hardware + Live Mode Dashboard

Priority: 🟢 Normal — maker/education community enablement.

Changes:

  • Create hardware/esp32/ with Arduino sketch for:
    • DHT22 (temperature + humidity)
    • MH-Z19B (CO₂ concentration)
    • VEML7700 or BH1750 (light intensity / PAR proxy)
    • WiFi + MQTT publish to habitat/sensor/* topics
  • Create hardware/README.md with:
    • Bill of materials (BOM) with approximate costs
    • Wiring diagrams (Fritzing or KiCad)
    • Sensor calibration guide
    • Mosquitto broker setup on Raspberry Pi
  • Add "Live Mode" tab to Streamlit dashboard:
    • Real-time MQTT sensor data vs. twin predictions
    • Visual diff overlay highlighting divergence
    • Alert when physical readings diverge from twin by >10%
  • Document the full HIL setup in docs/hardware_in_the_loop.md

Verification: End-to-end test with ESP32 hardware connected to the simulation via MQTT broker. Dashboard shows live sensor overlay.

Phase 3 — AI-Driven Autonomous Control (v0.4.0)

3A · Model Predictive Control (MPC) Optimizer

Priority: 🟢 Normal — intelligent ECLSS setpoint management.

Changes:

  • Create src/control/mpc.py with:
    • Objective function: minimize energy while maintaining O₂ > 19.5%, CO₂ < 5000 ppm, humidity within bounds
    • Prediction horizon: configurable (default 24h lookahead)
    • Control variables: PAR intensity, CDRA setpoint, dehumidifier setpoint
    • Solver: scipy.optimize.minimize (evaluate casadi if performance is insufficient)
  • Integrate MPC into the Simulation.step() loop as an optional controller
  • Add MPC toggle and parameter sliders to the Streamlit dashboard
  • Compare MPC-controlled vs. PID-controlled simulation runs
  • Write tests verifying MPC maintains safety margins under perturbation

Verification: MPC-controlled 72h simulation maintains all safety thresholds while using less "energy" (lower total PAR + CDRA duty cycle) than PID-only control.

3B · RL Agent + Mission Scenario Library

Priority: 🟢 Normal — long-duration adaptive control and educational scenarios.

Changes:

  • Create src/control/rl_agent.py with gymnasium-compatible environment wrapper:
    • Observation space: {O₂%, CO₂_ppm, humidity_kg, biomass_kg, Cᵢ, TTF, time_hours}
    • Action space: {PAR_intensity, CDRA_setpoint, dehumidifier_setpoint}
    • Reward: weighted sum of safety margins, energy efficiency, and crop yield
  • Train baseline PPO agent using stable-baselines3
  • Create src/scenarios/ with pre-built mission profiles:
    • mars_transit_150d.py — 150-day transit with crew sleep/wake cycling and periodic crop harvests
    • lunar_gateway_90d.py — 90-day lunar station with resupply events
    • sealed_room_72h.py — Short-duration sealed room (hobby/maker scenario)
  • Add scenario selector and RL agent toggle to the Streamlit dashboard
  • Monte Carlo uncertainty visualization (fan charts for O₂/CO₂ trajectories)
  • Export simulation results to CSV/Parquet

Verification: RL agent survives a 30-day simulation with injected perturbations without breaching safety thresholds.

Technology Stack Decisions

Need Tool Decision Rationale
CQY data source V-HAB polynomial matrices Adopted ✅ Primary source; 5×5 bivariate polynomials extracted
Crop parameters V-HAB PlantParameters.csv Adopted ✅ 9 crops extracted with full coefficient matrices
MQTT client paho-mqtt Adopt Already in pyproject.toml; production-grade
MPC solver scipy.optimize Adopt Already a dependency; sufficient for initial MPC
Nonlinear MPC casadi Evaluate later Only if scipy.optimize performance is insufficient
RL environment gymnasium Adopt De-facto standard; lightweight wrapper
RL agent stable-baselines3 Adopt Battle-tested PPO/SAC implementations
Data export pandas Adopt Already available; CSV/Parquet out of the box
Richer charts plotly Evaluate Optional; Streamlit native may suffice
Time-series DB InfluxDB Defer Only needed for persistent HIL storage
Target hardware ESP32 Prioritize Cheaper, Wi-Fi built-in, Arduino ecosystem
Linting ruff Adopted ✅ Fast, comprehensive; in CI pipeline
License MIT Adopted ✅ Maximum adoption; community standard for research tools

Contributing

See AGENTS.md for coding standards, testing requirements, and development workflow. All contributions must:

  1. Pass uv run pytest tests/ -v with zero failures
  2. Maintain coverage >80% (uv run pytest tests/ --cov=src)
  3. Include type hints compatible with strict mypy
  4. Follow conventional commit messages (feat:, fix:, refactor:, docs:, test:)
  5. Update CHANGELOG.md under [Unreleased] for every feature, fix, or refactor