KS
KstTai Space SystemsMission Autonomy Platform · KMAP
SIGNED IN AS
KMAP V1.2 · AI BACKEND + COLLABORATION + API + RBAC + AUDIT
KMAP / Mission Operations
V1.2 · AI-BACKED MVP EARTH OBSERVATION TWO-PERSON AUTHORIZATION RBAC

From mission intent to a verified, authorized plan.

KMAP compiles an operator's objective into constraints, generates and simulates candidate mission plans against a computed spacecraft state, verifies every plan against real engineering limits, and routes it through review and authorization before it becomes an operational plan. Nothing on this page is scripted — every number is produced by the mission compiler, a simulated-annealing planner, a simulator and a Monte Carlo verifier running in your browser. New in v1.0: role-based access control, a two-person review/authorize workflow, a live API console backed by the same engine, and a tamper-evident hash-chained audit trail — all persisted between visits.

Mission window
Spacecraft
Objective
Candidate activities

What KMAP does

01Understand
mission intent & requirements
02Model
spacecraft, payload & environment
03Plan
generate feasible schedules
04Optimize
trade mission value vs. resources
05Verify
test constraints in simulation
06Decide
recommend with human approval

KMAP mission workflow

MPQS selected plan
of max possible mission value captured
battery margin above required reserve
hard constraint violations

AI Mission Compiler · natural-language objective → structured requirement

Type a mission objective. KMAP's compiler parses it for hard constraints (battery reserve, thermal ceiling) and re-derives the requirement — no fixed template.

local heuristic

Planning engine · simulated annealing, not a fixed lookup

Each candidate plan is search-optimized: the planner starts from a heuristic schedule, then runs randomized local search — swapping activities in and out, re-simulating, and keeping improvements — until it converges.

Mission Control · Live What-If Analysis

Adjust priorities. KMAP's optimizer re-selects, re-simulates and re-verifies real candidate plans on every change.

Observation priority85
Battery conservatism70
Thermal conservatism75
KMAP recommendation: · computed score · mission value captured

Candidate mission plans · generated + simulated + verified this run

Selected plan timeline

00:00 ───────────────────────────────────────── 24:00
Battery % over mission window (dotted line = required reserve)

Spacecraft state at plan completion

Why KMAP selected this plan

Engineering verification · nominal

Monte Carlo verification · confidence under uncertainty

Nominal verification checks one trajectory. Monte Carlo verification re-simulates the selected plan hundreds of times with randomized energy/thermal noise and reports the probability each constraint actually holds.

Scenario testing

Re-run the simulator for the selected plan under a stressed environment model.

Review & collaboration · maker-checker before authorization

Engineering Reviewer signs off on the selected plan; only then can the Flight Director authorize it. Anyone can leave a note — comments are attributed and logged to the audit trail.

Human authorization gate

KMAP recommends a plan; the responsible operator remains in control. Authorization requires the Flight Director role and a completed engineering sign-off on this exact plan (two-person rule).

Status: Awaiting human approval.

Version history & audit log · hash-chained

not yet checked

API console · same engine, real requests

v1.0 exposes mission planning as an API. These calls run against the live in-browser engine (no external network) and return exactly what the production endpoint will return.

What KMAP is building toward

Mission Planningv0.3 mockup
Real Compiler + Plannerv0.4
Solver + Monte Carlo + Persistencev0.5
Collaboration + API + RBAC + Auditv1.0–v1.1
Real AI Backend (GapGPT)v1.2 — this build
Real-time state & ReplanningYear 2

v0.5 replaced the greedy planner with a search-based solver and added Monte Carlo confidence estimates and persistence. v1.0–v1.1 layered on role-based access control (Mission Operator, Engineering Reviewer, Flight Director, Viewer), a two-person maker-checker workflow, a documented API surface, expanded export formats and a hash-chained audit log. v1.2 replaces the compiler's local heuristic with an on-demand call to a real language model (via our own server-side proxy to GapGPT) — the "Refine with AI" button in the compiler panel — while keeping the local heuristic as an instant, offline-safe fallback. Next up (Year 2): real-time telemetry ingestion and closed-loop replanning.