NAVDRIFT-0 · Mission Control · ISRO SIH 2026 PS #26168

NAVDRIFT-0

ISRO SIH 2026 · PS #26168 · IDR System

NOMINAL

GNSS LOCKED

SIMULATION PS #26168

⚙

NavIC Constellation

7 SAT LOCKED

MET 00:00:00

Controls

Ground Truth: 📍 Load CSV

IDR System Modules

PS #26168

CAL ENG

VIB FILT

NHC

MAP MATCH

EKF/UKF

SNAP

ODO·AI

MAG/AHRS

DRIFT-FORMER · INT8/INT4 ONNX · 20ms CPU · 9-ch (baro+mag)

ZUPT

FED·LEARN

Non-Holonomic Constraints

Lateral vel

0.00

Vertical vel

0.00

Corrections0

Error removed0.00 m

Alignment & Vibration Filter

✓ LOCKED

PITCH

+3.2°

ROLL

-1.4°

YAW

0.1°

RAW IMU (noisy)

— km/h

AI FILTERED

— km/h

VIBRATION · raw (red) vs filtered (cyan)

Pothole events filtered0

WHEEL ODO

— km/h

AI IMU EST

— km/h

MAG heading—°

AHRS confidence—%

Event Timeline

0 steps

■ GNSS ■ Blackout ■ Tunnel ■ SNAP

Event Log

CARTOSAT-3 ROAD CONF

High ≥0.7

Med 0.4–0.7

Low <0.4

DR uncertainty band

GOOGLE MAPS API KEY

Position

—

Route

—

GNSS Mode

NavIC L5 Locked

DR Duration

0.0s

Ground Truth (RTK-GPS reference)

NAVDRIFT-0 · DRIFT-Former output

Raw IMU dead reckoning (no correction)

EKF baseline (classical filter)

HMM Map-Match ACTIVE

Live Telemetry

[SIM]

NAVDRIFT-0 ATE—m

EKF Baseline ATE—m

Uncertainty σ—m

Inference latency—ms

Steps / Outages0 / 0

ISRO Drift Compliance

Target <10%

0.0 %

Drift / Distance

ISRO limit: error/dist < 10%

✓ PASS

Dist: 0m · Err: 0.0m

GNSS ↔ DR Transition

ISRO req: seamless <100ms

0.4 ms

✓ PASS

Live Drift Race

NOMINAL

NAVDRIFT

— m

EKF

— m

Raw IMU

— m

NAVDRIFT saves vs IMU 0.0 m

NavIC Signal

NavIC L5 / S-Band

7 Satellites locked

PDOP

1.8

HDOP

1.2

VDOP

1.4

estimated (sim constellation)

Live Signals

Pos Error—

Uncert σ—

Speed—

Baro Alt—

9th ch—

DRIFTFormer Attention (illustrative — the exported ONNX graph does not emit attention weights)

50-frame window

t-50 peak: — t-0

Live Algorithm Race

lower=better

NAVDRIFT

—

EKF

—

RAW IMU

—

Inference Engine

Model: DRIFT-Former INT8 ONNX (14MB)
Runtime: ONNX Runtime 1.17 · CPU
Mode: Simulation

TRANSFORMER LAYERS · 4×8-HEAD · d=128

UKF State Vector

Covariance P

P[0][0] — Lat σ²

—

P[1][1] — Lon σ²

—

P[2][2] — Vel σ²

—

Trace(P)

—

ZUPT corrections: 0 ZUPT active: NO

Edge Deployment

INT8 ONNX target

💾

2.1

MB INT8

⚡

—

Infer Hz (live, ONNX on)

RAM footprint

34 MB

Target platform

ARM Cortex-A72

ISRO PS #26168 Compliance

WARMING…

50m blackout drift— warming up

1km tunnel ATE— warming up

Inference latency— warming up

Drift compliance— warming up

GNSS↔DR transition0.4 ms ✓

ZUPT corrections0

GNSS+INS Fusion

NavDrift

—

EKF

—

Raw IMU

low

Algorithm Benchmark

NAVDRIFT-0 ATE—m

EKF Baseline ATE—m

Raw IMU ATE—m

Adaptive RL (Q/R tuning)

Process noise Q—

Meas. noise R—

Residual buf size—

Sensor Fault Grid

Federated Learning (illustrative simulation — no real fleet is connected)

Round 0

Global ATE—m

Fleet size12 vehicles

TDP ETA—

Fleet Log

DRIFTFormer Attention (illustrative — not extracted from the real model)

axayazgxgygzspdbaroΔh

Attention Heatmap (4 heads) (illustrative)

📡

IMU PREPROCESSOR

Calibration · VibFilter · Baro(9th)

🧠

DRIFT-FORMER

Causal Transformer · RoPE · INT8/4

🔷

NAVIC VAE

Motion Prior Encoder

🗺️

MAP MATCH + NHC

Viterbi HMM · Non-Holonomic

⚡

GNSS+INS FUSION

EKF/UKF · SNAP · ODO·AI · 10Hz

📍

POSITION OUTPUT

SE(2) + Covariance · 10Hz