Attachment 2_TBD2_QA_V3_26NOV2025.pdf

ATTACHMENT 2: Questions and Answers V3 (Added Question 109 Only)

DARPA PS 26-01 (TBD2)

Q# Question Answer 1 Can the government provide the program-specific CUI guide? The solicitation referenced that it would be provided with the solicitation, but it is not on SAM.gov

The CUI guide will be provided upon offeror request. Please email darpa-ps-26-01@darpa.mil to request the CUI guide.

2 Participation from non-US persons: the security presentation at industry day included a reminder that CUI data cannot be shared with non-US persons. There was also a line in the slides that stated “data, analysis and results of space situational awareness is CUI” (transcribed from my written notes, may not be accurate). Please clarify the extent to which algorithms, data and results from this effort are expected to be CUI, as we need this information to inform our team structure.

Please reference the CUI Guide for the boundaries as to where the line is between UNCLASSIFIED and CUI. The CUI guide can be requested by emailing darpa-ps-26- 01@darpa.mil. Please see instructions in Amendment 02 to PS 26-01.

3 Will proposers day slides be provided?

Included JPL and LLNL?

Yes, slides will be made available.

4 Will the data sets be made available to the performers?

Yes, after award.

5 How soon will data be made available to performers?

Ground-based and synthetic images will be available to performers immediately after award.

Space-based images will be available later in the program.

6 Will synthetic data be CUI? Synthetic images will be unclassified. Real images from ground will be unclassified. Space-based images will be CUI.

7 Do you know the frame rate and noise level of the field data that will be provided?

Yes, the government will be providing that data to performers after award.

8 Will the data sets be divided into training data (labeled) and non-training data to support AI-based algorithms?

Performers will have to provide their own training data sets for AI-based algorithms. The government team may supply detailed physics models that performers can use to generate their own training data sets.

9 Do you have a specification on track accuracy and/or ability to forward project?

There are no specifications on track accuracy and/or ability to forward project.

10 What metadata will be provided with the data?

All nominal meta data will be supplied to performers upon contract award. Custom meta data can be requested, and the government may provide.

11 When should a performer expect technical details on the simulated and real datasets?

Will we receive this information prior to the submission of the abstracts, during the process leasing to oral presentations, or

Technical data on real and simulated data sets will be provided to performers after program kickoff will that only become available after program-start?

12 Should we anticipate the RSOs may not be on a fixed trajectory?

For synthetic images or real images with ground truth added, the government may include cases where the RSO maneuvers.

13 Persistent track implies custody. Does persistent track mean for the life of the mission?

TBD2 is focused on detecting and tracking RSOs in cislunar space. Once detections and tracks are established, the data volume is small enough that it can be transmitted to the ground for further analysis, to include establishing long term custody.

14 For more power consumption, what does

(T) and (O) mean?

Threshold and Objective

15 Data format: will data be provided as raw sensor data, or pre-processed image cubes?

As raw sensor data

16 Will characteristics of customer supplied data be provided, including iFOV, frame rate, resolution, etc.?

Yes, characteristics of government supplied data will be provided to performers.

17 Is the SRR a full SRR, or a pseudo SRR

(MVP)?

The program solicitation has been updated to define expectations for the design review.

18 Could you clarify "multi-sensor per platform designs are of interest as multiple telescopes reduce the number of pictures" from the PS? Why would more telescopes reduce the number of pictures?

Multiple telescopes in collapsed mode reduce integration time and since they have captured imagery synchronously, the images can be summed before being placed in the image cube.

19 Are the detection requirements near bright objects like the earth and moon? For example, do we need to detect 23 magnitude within 5 degrees of the moon?

There are no additional detection requirements near earth or moon. For task 2, this will be a consideration for the non-SEL1 employment scenario.

20 TBD2 metrics emphasize target detection, will state estimation accuracy be considered or is that out of scope?

Assuming "state estimation accuracy" refers to orbit determination, then the capability is of interest, but there is no program metric for it.

21 What are the constraints for the sensors and optics? Sensor resolution? FOV?

For Task 2, the government only sets a constraint on the diameter of the aperture. The other attributes of the sensor are part of the proposed design.

22 What is the range of object displacement speed in images?

Objects will be simulated on physical trajectories. The angular rates will depend on the distance and orientation of the sensor and target motion, but will be physically motivated given the target and sensor relative position and velocity.

23 For task 1, what should we assume for read out noise and background noise?

The real data from JPL will be 1.4 electrons per pixel, and the space data will be 0.8 electrons per pixel. Zodiacal light noise from realistic scenes will also be injected.

24 Please define integration time vs revisit time.

Integration time is the total time spent on the field of view when building a single data cube.

The revisit time is the max amount of time to scan all of cislunar space from SEL1. The max revisit time is 12 hours.

25 Is there a lower apparent magnitude target for ground based data, to account for atmospheric effects?

Yes, the ground-based data will not reach 23 magnitude.

26 How is the temporal extent of a data cube accounted for in the probability of false alarm?

Any time the performer reports the Pfa, it must meet the program metric. Integration time is not directly used in the assessment of false alarms.

27 A ground-based image cube is radically different from a space-based one. The PSF is time variable from the ground but very stable in space, meaning every star in a ground-based image cube changes its appearance every image. To run TBD on a ground-based cube, performers will have to PSF match them all, which is out of scope for space-based data cubes. Are you going to PSF match the ground-based data to reproduce the data one would acquire in space, or are performers going to have to include this treatment in proposed solutions?

In space the government will use exposures of a few seconds. On the ground, the PSF changes on a time scale of 10-100 msec, but when averaged over 1-3 s is moderately stable. There is no need to model time variability of PSF, but there is a need to model spatial variability of

PSF.

28 What is COTS or quasi-COTS? There are government funded advanced large sensors/FPAs…are they considered quasi

COTS?

Optics/sensors that are expected to be available at a TRL sufficient for transition by the end of the TBD2 period of performance are considered viable. The cost of customization, (flight qualification, etc.) should not dramatically exceed the "off the shelf" cost.

29 What qualifies as "quasi" COTS hardware?

What degree of customization is acceptable?

Optics/sensors that are expected to be available at a TRL sufficient for transition by the end of the TBD2 period of performance are considered viable. The cost of customization, (flight qualification, etc.) should not dramatically exceed the "off the shelf" cost.

30 What is meant by COTS and quasi-COTS? Optics/sensor expected to be available at a TRL sufficient for transition by the end of the TBD2 period of performance is considered viable. The cost of customization, (flight qualification, etc.)

should not dramatically exceed the "off the shelf" cost.

31 Is there a minimum TRL for the hardware design to be considered?

No, there is no hard limit on the minimum TRL, however, the payload design must be realistic in that it should be ready for being part of a payload at the conclusion of the TBD2 period of performance.

32 Does the team have any specific COTS hardware in view?

No, not at this time.

33 Who is the intended transition partner?

AFRL/RV, USSF, or JPL?

DARPA is engaging with multiple potential transition partners. The intended transition partner is not expected to impact TBD2 program metrics.

34 Are there assumptions/constraints to be made regarding target motion, such as orbital motion only, trajectory classes, maneuvering objects, min/max relative velocity, does customer data limit target motion? Are object discriminating ROQs anticipated?

Yes for orbital motion and transfer trajectories.

35 For algorithm evaluation, it was mentioned that both synthetic data cubes and real data frame stacks would be used. Is there any information at this time about the source of the real data? Interested in the aperture size, FOV, IFOV, focal plane, integration time.

All nominal meta data will be supplied to performers upon contract award. Custom meta data can be requested and may be provided.

36 Could you review your comments about potential GEO orbits that may be considered?

Deployment of sensors in GEO/beyond GEO orbits offer some capability of covering cislunar space. The solicitation mentions four areas of cislunar spaces that are of particular interest.

37 To what extent does payload design need to account for hardening and robustness of hardware?

While hardening and robustness of hardware is of importance for any space system, this is not the priority for this program.

38 Advantages, disadvantages, and restrictions on using COTS or rad-tolerant event-based sensors instead of CMOS?

If event-based sensors can meet program metrics and goals, they will be considered. However, the advantages will need to be justified in both the abstract and the orals.

39 Are CCD/CMOS the only possible sensors? Can we use our novel COTS sensors like Event Based Cameras or photon counting imagers?

Non CCD/CMOS sensors will be considered if they can meet program metrics and goals, but the advantages of these sensors over traditional sensors as well as their TRL will need to be convincingly justified.

40 Is the satellite platform stationary? For a deployment to SEL1, the satellite will be in a orbit around SEL1. The government is not prescribing if the system will be gimbaled or not.

41 Is the goal to observe the entire cislunar "sphere" or more of a "disk" from earth to moon? What about a smaller sphere say 50 km around the moon? Relevant timeline of hours for the processing?

The goal is to observe the cislunar sphere.

42 Can you provide SEL1 orbit parameters?

Can you better define cislunar volume?

How far above below the plane of the elliptic is included?

SEL1 is 1.5 million km from Earth between the Earth and the Sun. Cislunar volume is a sphere around the earth with a 400,000 km radius.

43 Can you give more info on SEL1 to satellite placement? What is allowable and what is not?

SEL1 is 1.5 million km from Earth between the Earth and the Sun. Employment will be an orbit around SEL1, other details are up to the proposer

44 Who determines the orbit(s)/locations for SEL1 deployment?

Exact orbit around SEL1 deployment is up to the proposer.

45 Why not a larger bus? The most likely employment of a TBD2 payload is as part of a rideshare, not a dedicated mission, which limits the size of the bus.

46 Will DARPA provide any bounds on relative rates for RSOs?

Yes. Cislunar orbits or orbits prior to achieving a cislunar orbit.

47 Will DARPA provide any specs for target astrometric accuracy?

Metrics for astrometric accuracy are not prescribed, but astrometric accuracy is one of the attributes of interest that will be taken into account in evaluations. The closer to photon limited accuracy for faint objects and to a fraction of arcsec, the better.

48 Will DARPA define any specs for stray light exclusion zones around earth and moon?

No, this is not an optics design effort.

49 Is it a TBD2 requirement to do on-board correlation and tracking of detections, across the 12 hour revisit gaps? Or does "tracking" only refer to the tracking across an image sequence to detect faint objects?

Once the object is detected, the data volume is small and can be transmitted to Earth.

50 What is the sensor aperture size of the quad telescope used for the real data set?

37 cm f/2.2 61Mpix cmos (a common COTS telescope/sensor)

51 Is the 0.5 m squared area total or per sensor?

0.5 m diameter is the per sensor limit for Task 2.

52 For the 0.5 m diameter limit on aperture diameter…how should this be interpreted in the case of non-circular or multi-aperture designs? Are non-circular apertures acceptable?

0.5 m is the diameter limit for a singular aperture. For non circular apertures, consider the constraint on aperture area pi*(0.5/2)^2

53 Does 0.5 m diameter apply to a single optic, or sum of all optics in a FPA? Is there a limit on the number of telescopes in an FPA?

0.5 m diameter is the limit for a single optic for Task 2. The use of multiple telescopes had advantages but will affect the total mass and volume of the system, so it should be part of the trade off study.

54 Do proposers need to address both the SEL1 and cislunar components or is just one fine?

For task 2, proposers will need to address payload designs for both the SEL1 and beyond GEO orbit employment scenarios.

55 Will a single or multiple algorithm proposal be expected/desired?

The government will not be prescribing the algorithmic approach.

56 Would interplanetary experience be weighted heavily on a team?

Only if the proposer demonstrates why it is of relevance.

57 Is task 1 more focused on detection, or tracking and orbital determination?

Task 1 is more focused on detection and tracking. Orbital determination can be performed on the ground.

58 Concern about radiation effects on optics that are collecting data for training . LEO based sensors will be less affected than images from GEO or higher orbits.

LLNL will make synthetic images as realistic as possible, to include radiation effect "squiggles".

59 Will algorithms just be run in simulation? The government will provide additional details on test capabilities and characteristics during program kickoff.

60 What level of detail on hardware is needed for combination with algorithms?

Exact hardware specifications are not needed at time of abstract or oral presentation.

61 Can you clarify the metric on integration time and processing time?

The compute needs to keep up with the data coming from the sensor(s). If processing takes longer than the integration time, but several integrations are processed in parallel then the processing may take longer than the processing.

The intent here is to say the system must not have a continuously growing backlog of processing tasks in a queue.

63 Are we expecting that performers will process one stack and then declare what is seen in that FOV?

Yes

64 Position = angular position, not cartesian position

The government is not expecting complete orbit determination to be performed in real-time on orbit. Position and velocity in the context of detection typically refers to angular position and velocity as viewed by the detection system.

65 Is the government open to using different telescope that are not in a FPA configuration, such as a splayed configuration?

The design of the optical system is part of the trade-space for performers to explore.

66 What size baffles will be needed for exclusion zones?

The design of the optical system is part of the trade-space for performers to explore in task 2, specifically for the non-SEL1 employment scenario.

67 For the trade study, how interested is the government in radiation effects, and different levels of rad hardened equipment?

While radiation tolerance is relevant to any space-based system, it is not the primary focus of this program.

68 Will algorithms be in real time on GPUs?

How critical will it be to plan for rad hard GPUs? Will porting be required?

The data is expected to be processed in near real time to prevent accumulation of unprocessed data. That said, the government is not requiring the use of a GPU. It is up to the performers to decide what kind of processor they require and the associated radiation tolerance.

69 Can proposers propose two linked/networked platforms in SEL1?

More than one platform at SEL1 is orthogonal to this program. The government is looking at metrics and the capabilities of a single platform.

Proposers are allowed to propose how additional platforms would enhance their capability. There is no metric in the program that is coupled with having multiple platforms at SEL1.

70 Is the government open to new developments, or do we want things that are proven, especially with regards to hardware?

Hardware that is expected to be available at a TRL sufficient for transition by the end of the TBD2 period of performance is considered viable. The cost of customization, (flight qualification, etc.) should not dramatically exceed the "off the shelf" cost.

71 What's the final deliverable of the algorithm? Does the algorithm need to be embedded in a specific language so that it can be deployed, or would a MatLab algorithm be appropriate?

The algorithms will have to be provided in an means that's executable via GPU, CPU, and/or FPGA; the government will not do any conversions.

72 Should proposers assume that the system will be read noise limited for the PFA or PD? Some cubes will be worse than others.

It depends on where the sensor is. But, PFA and PD are per cube.

73 Where did 300 TFLOPS come from? Value of 300 Tflops was the median value in a survey of existing algorithms processing a SEL1 scenario that had a 150 Mpix camera, 100 frames, 47,000 velocities, a total integration time of 700 seconds, and a 2 dB loss in SNR due to non-idealities.

74 Are the two employment scenarios alternatives to each other, or intended to be complimentary?

They are intended to be complementary

75 Is there a star catalog that will be recommended by the test and evaluation team?

GAIA3 and/or Pan-starrs

76 Would the government prefer traditional approaches or more AI/MIL-based approaches?

The government does not encourage or favor one specific approach over another, but is interested in approaches that can most likely meet program metrics and goals.

77 What theoretical framework should we use to define the limit as referenced in the PS:

"describe whether the proposed approach will be close to the theoretical limit of being able to detect a moving object in the same total integration time it would take to detect a stationary object." Should this be matched filter SNR, photo counting statistics, or other?

The theoretical limit refers to the integration time needed to detect a stationary object of given magnitude with a given sensor.

78 Is there a target percentage of theoretical efficiency DARPA considers acceptable?

80%? 90?

No, but the closer to 100% the better.

79 What approximate image dimensions and bit depth should be assumed?

In task1, this will be supplied along with the data. For task 2, it's the performer's choice.

80 Will the government T&E team simulate the conditions of the performers sensors for each performer during T&E?

The government will provide CMOS sensor data, and if a performer wants a different type of sensor data, the performer must provide sensor performance data to be simulated.

81 What are the specifications of the "government team hosted platform" for algorithm evaluation?

The government team will establish an Ethernet subnet in the Data Sciences Testbed (DST) for the edge computing evaluation effort. This will connect servers and edge devices that will be procured as needed for the performers. The general expectation is that the government will need and use Peripheral Component Interconnect Express (PCIe) devices.

82 If a new, innovative compute platform, coupled with our algorithms, is proposed, how will the government evaluate algorithm performance using a "common computer"?

TBD2 is focused on advancing the state of the art for signal processing algorithms that can operate on space-capable compute platforms, not advancing the state of the art of these compute platforms. Algorithms should be capable of running on existing processing technology.

83 Are we to share some code or dockers on LLNL testbed?

To be benchmarked, performers software will be run on the LLNL testbed.

84 Will LLNL's exaflop capability be made available for model training purposes?

No.

85 Are the threshold (600 W) and objective (300 W) power targets for the algorithm only, or the sum of the algorithm and payload?

The threshold and objective power targets are for the compute required to run the algorithms.

86 Does DARPA envision this being a classified effort or requiring a performance to have a facility clearance?

No, no facility clearance or security clearance is required for performance on the TBD2 program.

87 For the abstract phase, should the submission address both the SEL1 payload and beyond GEO payload, or focus just on

SEL1?

For abstract submissions, proposers only need to address SEL1 for both Task 1 and Task 2.

88 The evaluation of Pd/Pfa appears to be performed at the image stack/FoV level.

Please clarify how multi-scan/multi-look approaches that may revisit the same FoV multiple times before making a detection decision would be evaluated. It is understood that any multi-look approach must search the region of interest within the 12-hour revisit time.

Correct, Pd/Pfa are per one image cube.

Yes, multi-look approaches are allowed if the time to scan all of cislunar meets the revisit time metric.

89 Are the data cube images radiometrically accurate scenes as observed by the telescope or are they the outputs of a notional FPA? Is the FPA to be selected by DARPA or the vendor?

DARPA does not selects telescope or FPAs. Real images from ground or space will be output by real telescopes/FPAs, and metadata for those will be provided. Synthetic images will be created based on models of telescope and FPA.

90 The RFP page 6 identifies four possible missions for the placements of a few TBD2 sensors. Can the government confirm if the listed order is in priority order?

The list is not in priority order.

91 Can the government provide additional details on the range of apparent motions the proposed solution should expect?

No

92 What are the number of objects expected to be in the FOV for the test data?

From zero to 100s.

93 Is EO/IR okay for a sensor package, or just

EO?

IR is acceptable, but the advantages of using this approach must be made clear in the abstract submission.

94 Does the data need to be labeled with PNT? If so, is PNT data provided, or does that need to be part of the solution?

In task 1, pointing information is part of the meta data, which will be supplied with the image cubes. The performer is not expected to solve for orbit determination. Timing can be ignored.

95 For the two different sensor payloads, can two different processors be allowed?

Yes, each payload can have a different recommended processor.

96 Are performers able to restrict the search for faint objects to those whose trajectories are in or near the plane of the ecliptic?

No.

97 Are performers constrained to particular wavelengths, or are they free to select different wavelength bands for different aspects of the CONOPS?

Using different wavelengths is acceptable, but the advantages of using multiple wavelengths rather than the most common visible light window must be argued very well in the abstract.

98 Are FFRDCs/UARCs allowed as subcontractors to eligible primes?

No

99 Will CUI responses be accepted in order to provide applicable details?

Yes, CUI responses will be accepted

100 Do diagrams and figures count towards the 4 page limit?

Yes

101 Do citations count towards the 4 page limit?

No, citations will not count towards the 4 page limit for abstracts.

102 What are the restrictions on publication of academic research?

This effort is not expected to include fundamental research.

103 On page 6 of the PS there are two mentions of "bus" with respect to providing ROM weight, power, and cost.

Can you clarify that this is referring to the host spacecraft and how much detail DARPA is expecting on the bus/platform side or abstract and oral presentations?

A ROM is no longer required for abstracts;

please see Amendment 02 to PS 26-01.

104 Should cost ROM be included with then 4 pages or as a separate attachment?

A ROM is no longer required for abstracts;

please see Amendment 02 to PS 26-01.

105 What are the minimum data rights DARPA requires for the payload design to facilitate transition?

Please refer to the solicitation for information on data rights.

106 What format is preferred for the Price Summary?

Excel is preferred, but format is not prescribed in the program solicitation.

107 Can we submit multiple applications with different teams?

Only one abstract per prime CAGE code can be accepted.

108 As this will be using an Other Transaction Agreement, will cost sharing be required or not?

The OTA will not require cost-sharing unless the proposer is a traditional defense contractor who is not working with a non-traditional defense contractor participating in the program to a significant extent.

109 The Solicitation states FFRDCs are prohibited from proposing as performers;

does this also preclude FFRDCs performing as subcontractors?

As identified in Section 3.1(g) of the Solicitation: “University Affiliated Research Centers (UARCs), Federally Funded Research and Development Centers (FFRDCs), and government laboratories are prohibited from proposing as performers.” Accordingly, these entities are not permitted to propose as prime or subcontractor performers in support of the TBD2 Program.