Skip to content

IMPLEMENTATION_PLAN.md

Latest commit

 

History

History
162 lines (115 loc) · 5.31 KB

IMPLEMENTATION_PLAN.md

File metadata and controls

162 lines (115 loc) · 5.31 KB
Raw

MIME Implementation Plan

Phase 0 — Remaining Foundation (no physics textbook needed)

0A. Uncertainty Layer (src/mime/uncertainty/)

Depends on: Control layer (done) Blocks: Any closed-loop testing, B4/B5 benchmarks

  • UncertaintyModel ABC (proper base class with composability)
  • IdentityUncertainty (perfect sensing/actuation baseline)
  • ActuationUncertainty — frequency jitter, field inhomogeneity, pointing error, thermal drift
  • LocalisationUncertainty — Gaussian position noise, velocity noise, tracking dropouts, tracking_confidence field
  • ModelUncertainty — fractional noise on state fields (patient variability, fabrication tolerances)
  • ComposedUncertainty — stacks multiple models, model_a + model_b sugar

0B. Asset Schema (src/mime/schema/)

Depends on: Metadata (done), GeometrySource (done) Blocks: Registry integration, benchmark result attachment

  • MimeAssetSchema dataclass with all fields from ARCHITECTURE_PLAN.md §8
  • BenchmarkResult dataclass
  • mime_compliant property (compliance gate)
  • compliance_report() method
  • JSON serialisation (Phase 0–3 intermediate format before USD)
  • from_json() / to_json() round-trip

0C. Benchmark Stubs (src/mime/benchmarks/)

Depends on: Asset schema Blocks: Nothing directly — but establishes the test infrastructure for B0–B5

  • BenchmarkSuite class that discovers and runs registered benchmarks
  • B0–B5 as stub functions that raise NotImplementedError with clear messages about what's needed
  • Registration mechanism compatible with MADDENING's @verification_benchmark

0D. B0 Experimental Dataset Selection

Depends on: Nothing (literature research) Blocks: Phase 1 node design (constrains what RigidBodyNode must represent)

  • Confirm Rodenborn et al. (2013) as primary dataset
  • Document robot parameters, channel geometry, fluid properties
  • This is a research task, not a code task

Phase 1 — Core Physics Nodes (textbook required)

1A. External Magnetic Field Node

Depends on: MimeNode (done), Phase 0A complete Blocks: 1C, all magnetic actuation scenarios

  • Helmholtz coil (uniform field) model
  • Rotating permanent magnet (dipole) model
  • boundary_input_spec: frequency_hz, field_strength_mt, field_direction
  • These are the commandable_fields that ControlPolicy targets

1B. Magnetic Response Node

Depends on: 1A Blocks: 1C, B1

  • Permanent magnet response: T = m x B, F = grad(m . B)
  • Reads field_vector from 1A via edge, reads orientation from 1C
  • Outputs magnetic_torque, magnetic_force to 1C

1C. Rigid Body Node (6-DOF, overdamped Stokes regime)

Depends on: 1A, 1B Blocks: B0, B1, B2, everything else

  • Overdamped dynamics: velocity = R_T^{-1} * F_total, omega = R_R^{-1} * T_total
  • Quaternion orientation representation
  • Resistance tensor (sphere/prolate ellipsoid analytical, then RFT for helices)
  • Additive boundary inputs for forces/torques from multiple sources
  • Wall correction factors (Brenner) for confinement

1D. CSF Flow Node

Depends on: GeometrySource (done), 1C Blocks: B0, B2, B4

  • Start with analytical Stokes drag (no resolved flow field) — sufficient for B0, B2
  • Pulsatile component via Womersley analytical profiles in cylindrical geometry
  • Later: IB-LBM for resolved flow
  • Bidirectional coupling: receives robot position/velocity, returns drag force/torque

1E. Phase Tracking Node

Depends on: 1A, 1B Blocks: B1, B5

  • Observer node (not physics)
  • Reads orientation from 1B, field rotation from 1A
  • Computes phase_error = angle between magnetic moment and external field
  • Step-out detection: phase_error > pi/2

1F. Benchmarks B0, B1, B2

Depends on: 1A–1E all complete

  • B0: trajectory comparison against Rodenborn et al. experimental data
  • B1: step-out frequency vs. regularised Stokeslet reference
  • B2: drag force vs. Stokes law at Re < 0.1

Phase 2 — Drug Delivery + Realistic Environment

2A. Drug Release Node

Depends on: 1C

  • First-order kinetics, Higuchi model, Korsmeyer-Peppas
  • Trigger mechanisms (passive, pH, magnetic, acoustic)

2B. Concentration Diffusion Node

Depends on: 1D, 2A, GeometrySource

  • Advection-diffusion equation
  • Potentially wraps MADDENING's HeatNode (Mode 1)

2C. MRI Signal Formation Node

Depends on: 1B, 1C

  • Susceptibility artefact model
  • Feeds into UncertaintyModel for realistic position noise

2D. Flexible Body Node

Depends on: 1C

  • Cosserat rod / discrete elastic rods for flagellar robots

2E. Surface Contact Node

Depends on: 1C, GeometrySource

  • Penalty-based contact, adhesion models
  • Near-wall hydrodynamic corrections

2F. Benchmark B3

Depends on: 2A, 2B

2G. Additional nodes (parallel)

  • Non-Newtonian Rheology, Acoustic nodes, etc.

Phase 3 — Closed-Loop + Robustness

3A. StepOutDetector Feedback Policy

Depends on: 1E, control layer (done)

3B. State Estimator (EKF for robot pose)

Depends on: 2C, uncertainty layer

3C. Benchmarks B4 (T1, T2, T3), B5

Depends on: 3A, 3B, Neurobotika mesh (external)

Phase 4 — MICROBOTICA Integration

4A. USD Asset Serialisation

Depends on: Asset schema, stable node APIs

4B. MICROBOTICA Desktop Integration

Depends on: 4A, MICROBOTICA Phase 0

4C. Registry API + Leaderboard

Depends on: 4B