TBD2_DARPA_proposers_day_briefing.pdf

Track at Big Distances with Track-Before-Detect

(TBD2)

Dr. Stefano Galli Program Manager, DARPA/STO

Proposers Day

November 4, 2025

Distribution Statement A: Approved for public release. Distribution unlimited.

Agenda

Distribution Statement A: Approved for public release. Distribution unlimited.

Time Topic Speaker 0900 – 0930 Check-In Registration Desk 0930 – 0940 Welcome / Security Overview Dr. Stefano Galli / Linnea Goldberg 0940 – 1010 TBD2 Program Overview Dr. Stefano Galli 1010 – 1025 JPL Michael Shao 1030 – 1045 LLNL Nate Golovich 1045 – 1100 Contracting Overview Rana Movahed 1100 – 1110 DARPAConnect Sana Hoda 1110 – 1210 Lunch 1210 – 1255 Q&A 1300 – 1700 One-on-ones (15 min each) Check-in at registration desk

TBD2 Proposer’s Day – Tuesday, November 4th

Track at Big Distances with Track-Before-Detect

(TBD2)

Dr. Stefano Galli Program Manager, DARPA/STO

Proposers Day

November 4, 2025

TBD2 Overview

36k km

384k km

Technical Hypothesis Combining signal processing algorithms with onboard processing and commercially available sensors enables SSA at cislunar distances and beyond, allowing detection and tracking of at least 1 m wide RSOs at 2 Gm and revisit times of less than 12 hours

Vision Enable continuous, uncued detection and tracking of objects in the extended cislunar space on a relevant timescale to prevent surprise

GEO

TBD2 expands SSA beyond GEO to cislunar space

Cislunar volume: 237 peta-km3

( 1,200x GEO volume)

LEO

GEO volume: 0.2 peta-km3

Gm: 1 Million km GEO: Geosynchronous Orbit LEO: Low Earth orbit SDA: Space Domain Awareness RSO: Resident Space Object

Continuous tracking can be addressed in two ways

SEL1

1.5 Gm

Sensing at SEL1

• Sun always behind sensor

• Small solid angle subtends Cislunar

Sensing at 5x-20x GEO

• Multiple sensors in various orbits

• Diversity of sensor positions minimizes downtime

SEL1: Sun Earth Lagrange Point 1

What if:

• We could persistently track 1 m RSOs at 2 Gm and scan all of cislunar in less than 12 hours?

• We could use a few additional sensors to track smaller RSOs in some important parts of cislunar space?

• We could do all this with a revisit time of less than 12 hours?

SOA: few exquisite assets, downtime due to sun, and SNR too low for detecting reliably dim objects

TBD2 will demonstrate that the program goals can be achieved with COTS sensors provided we use detection and tracking algorithms that:

• Operate reliably at very low SNR; and

• Have computational complexity low enough for on-board processing

SNR: Signal-to-Noise Ratio EML1: Earth – Moon Lagrangian 1 SST: Space Surveillance Telescope GEODSS: Ground-Based Electro-Optical Deep Space Surveillance

Challenge: reduce computational complexity of signal processing algorithms to run on low SWAP platform

While the family of Track-Before-Detect algorithms is well suited to operate on stacks of low SNR images, their computational complexity is too high for low SWAP platforms 0 5 10 15 20 25 30 35 40 45 50

Revisit Time (hours)

10 -1

10 0

10 1

M in im um

D et ec ta bl e

R

SO

S iz e (m

RSO at a distance of 10xGEO (Moon orbit)

SST (Ground)

GEODSS (Ground)

Oracle (EML1 2027) 𝒎𝒎𝒗𝒗 = 𝟏𝟏𝟏𝟏 𝒎𝒎𝒗𝒗 = 𝟐𝟐𝟐𝟐 𝒎𝒎𝒗𝒗 = 𝟏𝟏𝟏𝟏

SOA: State of the Art

Algorithmic complexity vis-à-vis available compute

Latest technological advances in rad-hard computing are not enough to support the computing needs of cislunar SSA, TBD2 is necessary to close to gap

CP: computational power

Quantitative shift of power/mass vs revisit time

Only with TBD2 can we achieve both operationally relevant revisit times and reasonable mass

TBD2

From 62 days revisit time to 12 hours

Payload mass reduced by 80x

Distribution Statement A: Approved for public release. Distribution unlimited.

10 1 10 2 10 3

Revisit Time (hours)

10 2

10 3

10 4

10 5

C om pu te

S en so r M as s (k g)

10 2

10 3

10 4

10 5

C om pu te a nd

S en so r P ow er

(W

SOA vs. TBD2 Benefits for SEL1

SOA Mass Operating Line

TBD2 Mass Operating Point

Delta IV Mass Capability

Starship Mass Capability

Ariane VI Mass Capability

SOA Power Operating Line

TBD2 Power Operating Point

Signal Processing

• Performers will develop low complexity algorithms capable of running via onboard compute capabilities

• Performers will be provided with image cubes, including:

• Real cislunar images taken from space by government partners

• Real images taken from ground with a quad telescope

• Partially synthetic images

• Synthetic images

• Some image cubes will have fake RSOs inserted

• Signal processing performance will be evaluated on government-provided compute platforms at the midpoint and end of program

Program plan

Distribution Statement A: Approved for public release. Distribution unlimited.

Payload Design

• Performers will develop two payload designs:

• For placement at SEL1 (2 Gm distance, low angular velocity)

• For placement in a beyond GEO/cislunar orbit (0.4 Gm distance, higher angular velocity)

• Includes sensor, compute platform, and algorithms

• Evaluated via System Requirements Review (SRR) near end of program

Program Schedule

Program Kickoff

Midterm evaluation

Task 2: Payload design trade study

Task 1: Track-Before-Detect algorithms

Final Evaluation And SRR

15 months

Metrics and Constraints

Metric Name Single Phase M1 Mag-Range/Size 23 or better (1 m @ 2 Gm, 0.1 m @ 0.2 Gm) M2 Revisit time <12 hours for single asset at SEL1 M3 Max power consumption (T) 600 W; (O) 300 W M4 Probability of Detection > 95% M5 Probability of False Alarm <1%

Constraints Albedo: ≤ 20% Processing time ≤ total integration time Optic aperture size ≤ 0.5 m diameter

TBD2 government team

Air Force Research Laboratory

• Agreements Officer Representative (AOR)

• Provide subject matter expertise from Oracle-M and Oracle-P

• Provide space-based imagery from Oracle-M

• TBD2 will transition best algorithms to Oracle-P for testing

Lawrence Livermore National Lab

• Performer evaluation and selection support

• Architecture trade studies support

• Generation of synthetic image cubes with RSOs

• Hosted compute platform and algorithm benchmarking

Jet Propulsion Lab

• Performer evaluation and selection support

• Architecture trade studies support

• Generation of real data and real data with fake objects

• Algorithm evaluation for SSA

Security

One-on-one schedule www.darpa.mil

15Distribution Statement A: Approved for public release. Distribution unlimited.