DARPA-PS-27-06.pdf

Program Solicitation

Strategic Technology Office (STO)

Unison

DARPA-PS-27-06

2 November 2026

Table of Contents

1. Program Overview

1.1. Background and Motivation

1.2. Program Structure

1.3. Technical Challenges

1.3.1. Unison Technical Area 1 – Joint Co-adaptation

1.3.2. Unison Technical Area 2 – Modeling and Simulation Environment

1.4. Program Metrics

1.4.1. TA-1 Metrics

1.4.1.1. TA-1 Technical Performance Measures

1.4.1.2. TA-1 Constraints

1.4.2. TA-2 Metrics

1.4.3. TA-2 Constraints

1.4.4. Government-Furnished Software Benchmark

1.5. Hypothetical Example Scenario – Civilian Search and Rescue

1.6. Program Test Events

1.6.1. Phase 1 (15 months)

1.6.2. Phase 2 (15 months)

1.7. Government Furnished Information and Software

1.8. Kickoff Meetings

1.9. Agile Development, Monthly Deliveries, and Transition Engagement

1.10. Final Report

1.11. Intellectual Property (IP)/Data Rights

2. PS AUTHORITY

2.1. PS Procedure

2.2. TA-1 Selection Process

2.3. TA-2 Selection Process

3. GUIDELINES FOR ABSTRACTS, ORAL PRESENTATIONS, DEMONSTRATIONS, AND

PROPOSALS

3.1. General Guidelines

3.2. TA-1 Abstracts Gate 1 and Gate 2 Content

3.3. TA-1 Abstracts and Presentation Slides (Gate 1) – Process and Basis of Evaluation

3.4. TA-1 Oral Presentation and Q&A (Gate 2) – Process and Basis of Evaluation

3.5. TA-1 Written Proposals Content

3.6. TA-1 Written Proposal – Process and Basis of Evaluation

3.7. TA-2 Gate 1 Content

3.8. TA-2 Abstracts and Video – Process and Basis of Evaluation

3.9. TA-2 Written Proposals and Baseline Test

3.9.1. Baseline Test Demonstration

3.9.2. Written Proposals

3.10. TA-2 Written Proposal and Baseline Test Demonstration – Process and Basis of Evaluation

4. AWARDS

4.1. General Guidelines

4.2. Controlled Unclassified Information (CUI) and Controlled Technical Information (CTI) on Non- DoD Information Systems

4.3. Representations and Certifications

4.4. Competition Sensitive Information

4.5. Procurement Integrity Act (PIA)

5. PS DEFINITIONS

6. ACRONYMS

7. GLOSSARY

8. Appendix 1: Hypothetical Mission Scenario – Civilian Search in Advance of Rescue Crew

8.1. Hypothetical Mission Narrative

8.2. Hypothetical Scenario’s Mission Parameters

8.3. Hypothetical Scenario’s Notional Vehicle and Sensor Parameters

8.4. Hypothetical Scenario’s Communications Parameters

8.5. Hypothetical Scenario’s Mission Rules and Success Criteria

8.6. Hypothetical Scenario’s Expected Autonomous Collaboration

9. Appendix 2: DARPA-PS-27-06 CUI ADDENDUM REQUEST FORM

To be provided separately by Government with invitation to submit a Written Proposal:

Attachment 1 – Unison Model Other Transaction for Prototype Agreement Attachment 2 – Task Description Document Attachment 3 – DARPA Standard Cost Proposal Template Attachment 4 – Schedule of Milestones and Payments Attachment 5 – Property / Equipment Purchases Attachment 6 – Wide Area Work Flow (WAWF) Instructions Attachment 7 – Data Rights Assertions Attachment 8 – Value Assessment Questions for Proposers Attachment 9 – Performer Attestation of Nontraditional Defense Contractor Status

PROGRAM SOLICITATION OVERVIEW INFORMATION

• Federal Agency Name – Defense Advanced Research Projects Agency (DARPA), Strategic Technology Office (STO)

• Funding Opportunity Title – Unison

• Announcement Type – Initial Announcement

• Funding Opportunity Number – DARPA-PS-27-06

• Dates - All times are Eastern Time (ET) o Posting Date: 2 November 2026 o Proposers Day: 5 November 2026 o Addendum Request Date: 12 November 2026 o Questions Submittal Closed: 16 November 2026 o Abstracts Due Date and Time:

▪ TA-1 Abstracts and Slides: 23 November 2026 at 5:00PM ET

▪ TA-2 Abstracts and Video: 4 December 2026 at 5:00PM ET o TA-1 Oral Presentation and Q&A: ~7 January 2027 o TA-1 Written Proposal: ~early February 2027 o TA-2 Written Proposal and Baseline Test: ~late January 2027

• The Defense Advanced Research Projects Agency (DARPA) is soliciting innovative proposals to address challenges in the following technical area: developing collaborative autonomy that jointly co-adapts autonomy and communication functions in order to maximize mission success under increasing mission complexity and operational variability.

• Multiple awards are anticipated.

• Types of instruments that may be awarded – Other Transaction Agreement (OTA) for Prototypes

• Agency Contact The Solicitation Coordinator for this effort can be reached at: Unison@darpa.mil or via U.S.

postal service to:

DARPA/STO

ATTN: DARPA-PS-27-06

675 North Randolph Street Arlington, VA 22203-2114

• Security Requirement for Prime Proposers: There will be work conducted on Unison at the Controlled Unclassified Information (CUI) level. All individuals creating or accessing information on Unison agree to protect CUI in accordance with DoD Instructions 5200.48 and National Institute of Standards and Technology (NIST) Special Publication 800-171 Rev 2 Protecting Controlled Unclassified Information in Nonfederal Systems and Organizations.

Proposer must be compliant with NIST Special Publication 800-171 Rev 2 at onset of the program.

No foreign nationals (also referred to as foreign persons) shall be permitted to perform any work on this effort, either as prime contractor personnel or as subcontractor personnel at any tier. For the purposes of this solicitation, a foreign national is any individual who is not a U.S.

citizen or a U.S. lawful permanent resident (Green Card holder).

This restriction applies to all personnel who have access to technical data, software, CUI, or who will contribute to the technical performance of the contract. The prime contractor shall flow this restriction down to all subcontractors and shall ensure compliance.

• CUI Addendum:

The effort solicited by this Program Solicitation (PS) is Controlled Unclassified Information (CUI). A formal request for the CUI addendum must be submitted by completing the DARPA-PS-27-06 CUI Addendum Request Form (Appendix 2) and emailing it to Unison@darpa.mil with the subject line “Request DARPA-PS-27-06”.

Prior to releasing the CUI addendum, the Government will verify that the requesting entity meets the CMMC Level 2 (Self-Assessment) requirements for this solicitation. This verification will be performed by checking the Supplier Performance Risk System (SPRS) using the CMMC Unique Identifier (UID) provided on the request form.

To receive the CUI Addendum electronically, the requester’s UID must be associated with a current CMMC Level 2 (Self-Assessment) status (posted as "Final" or "Conditional").

Requesters that do not meet this CMMC requirement in SPRS will receive physical copies mailed via USPS Registered Mail or Express Mail to a verified business facility. If physical copies are mailed, proposers must explicitly certify that they will handle, store, and destroy the physical documents in accordance with federal CUI standards.

Once DARPA confirms CMMC status, additional materials will be sent to requestor via DOD Safe or mail.

The CUI addendum contains the following materials:

o Controlled Unclassified Information (CUI) Guide. All proposers to either TA-1 or

TA-2 must submit a request for the CUI Guide by the Addendum Request Date in order to submit an abstract.

o Government-furnished reference autonomy software based on the DARPA Contested Operations in Denied Environments (CODE) program. Includes software and supporting documentation. Proposers to TA-2 must submit a request for CODE by the Addendum Request Date in order to submit an abstract and video for TA-2.

PROGRAM SOLICITATION

Defense Advanced Research Projects Agency (DARPA)

Unison

1. Program Overview The Defense Advanced Research Projects Agency (DARPA) is soliciting innovative proposals for Unison, a program to develop and evaluate prototype collaborative-autonomy approaches that jointly adapt team decision-making and communications during mission execution. Unison will investigate approaches that enable autonomous teams to modify tactics, behaviors, coordination relationships, and information exchange as mission complexity and operational variability increase and communications become constrained and/or degraded (see Section 7, Glossary, for program-specific definitions of terms in italics).

The core hypothesis of Unison is that a unified execution-time reasoning approach to action planning and communications planning can improve the effectiveness, scalability, and resilience of collaborative autonomy in dynamic, uncertain, and resource-constrained environments. Unison seeks to preserve mission performance while reducing communication dependence and bounding the reasoning burden imposed on individual agents.

Unison frames collaborative autonomy as a multi-agent decision problem in which agents use local observations, mission context, relevance, and mission objectives to select a policy that jointly defines physical actions and communications actions. At each decision point, the team must determine what actions to take, which coordination relationships must be maintained, and what information must be exchanged to preserve the coordination required for mission success. Unison will investigate approaches across teams of tens to hundreds of agents operating in representative collaborative, multi-domain mission scenarios.

DARPA seeks proposals in two Technical Areas (TAs):

• TA-1, Joint Co-Adaptation: development of collaborative-autonomy planning and execution software prototype approaches that jointly adapt autonomy and communications during mission execution.

• TA-2, Modeling and Simulation Environment: development of a Faster-Than-Real-Time (FTRT) modeling and simulation prototype environment to train, integrate, evaluate, and stress-test TA-1 approaches under representative mission conditions.

TA-1 and TA-2 are interdependent: TA-1 will develop the co-adaptation approaches, while TA-2 will provide the common faster-than-real-time (FTRT) training and evaluation environment needed to develop, train, and measure TA-1 solutions’ mission performance, computational burden, communications demand, and resilience under controlled, repeatable conditions.

1.1. Background and Motivation

Current autonomous systems face challenges in their ability to adapt and re-plan as mission complexity and operational variability increase, particularly when they depend on reliable, high-bandwidth communications. A primary issue is the separation of autonomy from communications in system design and adaptation: autonomy determines tactics, behaviors, and coordinated decisions, while the communications layer determines how data is transported across the network of systems.

The autonomy layer generally does not adapt to changes in communication availability, while the communications layer generally does not account for mission value or coordination consequences of the information they exchange.

Figure 1 illustrates the current state of practice along two dimensions: adaptation of autonomy and adaptation of communications. The figure identifies three current approaches and the unoccupied upper-right quadrant that Unison will investigate. Not of interest are solutions that exclusively occupy quadrants one, two or three.

• Quadrant 1 (lower-right): 6G semantic communications

Semantic communications approaches adapt what information is transmitted based on receiver relevance or mission value, potentially reducing bit transmission while preserving information useful to the receiver. However, these approaches generally hold autonomy policies fixed; they do not adapt team tactics, behaviors, or coordination decisions in response to the same operational and/or network conditions changes.

Figure 1: Current architectures blind tactical adaptability from network adaptation.

Unison occupies the critical empty Q4 where both comms and autonomy functions co-adapt

• Quadrant 2 (lower-left): Centralized “fire-and-forget” systems

Operating on centralized offline plans and pre-programmed execution rules (e.g., pre-programmed one-way attack drones’ strike salvos, commercial drone shows, etc.), these systems can function without communication. However, their pre-programmed logic assumes low operational variability.

Typical approaches like “finite state machines” operate based on planning time assumptions with limited ability to adapt to execution time variabilities. When the pre-programmed rules are invalid and the current mission context requires re-planning, this approach to autonomy is challenged to adapt and re-plan, impacting mission performance.

• Quadrant 3 (upper-left): “Communication-dependent” distributed autonomous systems

These systems can dynamically adapt tactics and behaviors during execution but often depend on frequent information exchange to maintain the shared or sufficiently consistent team state required for coordinated action. As team size, observation volume, and mission dynamics increase, the communications burden needed to sustain that coordination can grow substantially. When communications are degraded or unavailable, agent beliefs can become incomplete or stale, impacting coordinated action.

Taken together, Quadrants 2 and 3 illustrate a central challenge of current approaches: systems may be less dependent on communications but unable to re-plan effectively, or they may adapt to changing conditions but remain dependent on continued information exchange to maintain coordination. In either case, the system risks impacting effective team coordination as operational conditions depart from planning assumptions or communications degrade.

• Quadrant 4 (upper-right): Joint adaptation of autonomy and communications

The unoccupied quadrant requires both the autonomy policy and the communication policy to co-adapt during mission execution. Such joint policies must determine:

• which agents need to coordinate;

• which states, observations, actions, and future conditions are relevant to that coordination;

and

• what information must be exchanged to preserve coordinated action.

For a Quadrant 4 solution, in which autonomy and communications jointly adapt during mission execution, effective mission execution is contingent on maintaining coordinated action across the team. Maintaining coordination requires agents to either exchange information to reduce uncertainty about relevant teammate and environment states or to infer those states locally. Frequent information exchange increases communications demand; when communication is limited, the larger uncertainty each agent must manage expands to a decentralized partially observable Markov decision process (Dec-POMDP)-style reasoning problem, increasing reasoning burden. This relationship creates a trade space between reasoning (computation) burden and communications burden. As mission complexity and operational variability increase, both the information needed to sustain coordination, and the uncertainty an agent must manage when information is unavailable, also increase. Existing approaches generally optimize one side of this trade space while treating the other as fixed. Figure 2 illustrates the need to jointly manage reasoning burden and information exchange.

Unison will investigate joint action–communications policies initialized at planning time using mission objectives, constraints, and available resources, but executed and adapted during mission execution. The program will evaluate the ability of these policies to determine what each agent must reason over, which agents and decisions must remain coordinated, and what information must be exchanged to preserve coordinated mission progress. Rather than requiring continuous coordination across the full team state, Unison will adapt reasoning scope and communications dependencies according to mission context and relevance.

Figure 2: Unison must balance reasoning burden and communication burden

1.2. Program Structure

Unison is a 30-month program organized into two phases: Phase 1 (15 months), followed by a Phase 2 (15 months). The program includes TA-1, Joint Co-Adaptation, and TA-2, FTRT Modeling and Simulation Environment. Both TAs support the program objective of co-adaptation of autonomy and communications. The 30-month schedule is shown in Figure 3.

Figure 3: Unison Program Schedule

DARPA anticipates multiple awards to TA-1 proposers for Phase 1 with a downselect in Phase 2.

TA-1 performers will demonstrate progress through five test events (TE) of increasing mission complexity and operational variability. Two test events will occur during Phase 1 and three test events will occur in Phase 2. TA-1 performers will also conduct self-evaluations and provide progress reports to the Government team.

DARPA anticipates multiple awards to TA-2 proposers for Phase 1. TA-2 performers will demonstrate an initial testbed capability by Month 4 against the TA-2 metrics and participate in recurring technical exchanges with the Government team. DARPA will downselect to one TA-2 performer following the initial testbed evaluation at Month 4. The selected TA-2 performer will provide the common FTRT training, integration, and evaluation environment for TA-1 solutions for the remainder of Phase 1, with an option to continue into Phase 2.

DARPA will provide mission scenarios for TA-1 and TA-2 performers (see Appendix 1). The Government will evaluate performer progress during each test event against program objectives, technical-area requirements, and program metrics, as described in Section 1.4. Additionally, the Government will also provide a Government-Off-The-Shelf (GOTS) autonomy solution that will be used for benchmarking TA-1 performance, as Government Furnished Software (see Section 1.7).

This solicitation includes both TA-1 and TA-2 for both Phases 1 and 2. Proposers are allowed to submit abstracts to both TAs; however, they must submit separate abstracts for each TA, with each abstract addressing only a single TA. If a proposer team submits abstracts for both TAs and is invited to submit full written proposals, the proposer team must choose only one TA to submit a full proposal since proposers cannot be a performer for both TAs. Proposers may submit up to one (1) abstract per TA. Proposers with different Unique Entity Identifier (UEI) registered in the System for Award Management (SAM.gov) and associated Commercial and Government Entity (CAGE) code are considered distinct proposers.

TA-1 proposers are expected to submit a Base Phase 1 proposal with an option for Phase 2. TA-2 proposers are expected to submit a base proposal through Month 4 (Baseline test), an option for the remainder of Phase 1, and an option for Phase 2.

1.3. Technical Challenges

Unison presents two coupled technical challenges: developing collaborative-autonomy approaches that jointly adapt autonomy and communications during mission execution, and developing a simulation environment capable of training, testing, and stressing those across various civilian and military mission scenarios with increasing mission complexity and operational variability. TA-1, Joint Co-Adaptation, and TA-2, Modeling and Simulation Environment, each address a distinct part of these challenges.

TA-1’s overall challenge is to enable autonomous teams to make coordinated decisions that achieve mission objectives as mission complexity and operational variability increase, while keeping per-agent reasoning and communications demand within specified bounds. Meeting this challenge requires dynamically managing the trade space among computational tractability, communication efficiency, and operational resiliency, as described in Section 1.3.1.

TA-2’s overall challenge is to develop a statistically robust, FTRT modeling and simulation environment capable of training and evaluating TA-1 solutions at operationally relevant scales. The environment must support repeated execution of black-boxed TA-1 (or Government-Furnished software) solutions against common mission scenarios while enabling controlled variation in mission conditions, communications conditions, and other factors material to TA-1 decision-making, as described in Section 1.3.2. It is expected that results proven in the simulated environment will lay a solid foundation for eventual future live tests. For example, realistic scene generation and sensor data (image) construction should reflect performance of real sensors.

TA-1 and TA-2 performers will participate in a Government-led Integrated Product Team (IPT) to define and document a standardized black-box interface between TA-1 autonomy solutions and the TA-2 testbed. The interface will provide the instrumentation and data capture needed to support execution, training, calculation of Government-defined performance metrics, and diagnosis of solution behavior during test events. The Government expects the API to be ready for TA-1 implementation by Month 4. Drafts of the API will be available in the interim to help TA-1 performers prep for final release.

The two technical areas are interdependent. TA-1 cannot credibly demonstrate the co-adaption policy across various mission scenarios with increasing mission complexity and operational variability without a valid TA-2 training and evaluation environment. TA-2, in turn, must focus its simulation fidelity, orchestration framework, and measurement approach on the factors that materially affect TA-1 decisions. Together, the Technical Areas must demonstrate that joint autonomy– communications co-adaptation can reduce computational and communications burdens while retaining mission success as missions become more complex, environments more variable, and adversary behavior becomes more adaptive. Figure 4 illustrates the relationship between TA-1 and

TA-2.

Figure 4: TA-1 and TA-2 architectural dependencies.

The components within each box are illustrative examples (and should not be considered as required) of the types of functions expected.

1.3.1. Unison Technical Area 1 – Joint Co-adaptation

Collaborative autonomy can be modeled as a multi-agent decision problem in which agents select physical and communication actions based on local observations and a belief about the joint team and environment state. The objective is to sustain the coordination required to advance shared mission progress towards success. At each decision point, agents must determine both what physical actions to take and what information to exchange with relevant teammates to preserve coordinated mission progress.

At the team level, this decision process can be represented conceptually as a joint action– communication policy:

TA-1

TA-2

TA-2

TA-1/2

𝜋:𝑏𝑡↦(𝑎𝑡,𝑐𝑡) where 𝑏𝑡 is the team’s current belief about the joint state, 𝑎𝑡 is the set of joint physical actions, and 𝑐𝑡 represents communication actions at time t. Unison seeks policies that maximize mission success while dynamically balancing three coupled technical dimensions: computational tractability, communication efficiency, and operational resilience.

A successful TA-1 solution will engineer, execute, and continuously adapt joint action– communications policies that meet mission objectives provided at planning time while dynamically balancing these three dimensions:

• Computational Tractability: Bound the effective per-agent reasoning problem as mission complexity and operational variability increase. Higher values of computational tractability indicate lower "Reasoning Burden"

• Communications Efficiency: Localize coordination and reduce explicit information exchange while maintaining the information necessary for coordinated mission progress. Higher values indicate lower "Communication Burden".

• Operational Resilience: Sustain required mission performance as operational variability increases. Higher values indicate ability to adapt to (tolerate) uncertainty.

These dimensions are coupled. Reducing information exchange can improve communications efficiency but increase local uncertainty, increasing reasoning burden and reducing computational tractability. Conversely, increasing information exchange can reduce local uncertainty, lowering reasoning burden and improving computational tractability, but at the cost of lower communications efficiency. Localizing coordination can increase both computational tractability and communications efficiency, but excessive localization can decrease operational resilience when broader awareness is needed. TA-1 solutions should therefore dynamically adapt their operating point across these dimensions during execution rather than optimize any one dimension independently.

Proposers will be evaluated based on the TA-1 metrics outlined in Section 1.4.1. These metrics assess computational tractability, communications efficiency, operational resilience, joint co-adaptation, and mission success. The metrics are intended to assess both whether a solution meets required performance bounds and whether it can adapt across the trade space as mission conditions change.

TA-1 proposers should provide evidence of their ability to meet the TA-1 technical challenges and explain why their approach is expected to achieve program objectives, by detailing:

a) Algorithmic/Technological Framework: Describe the technical insights, architectures, and methods used to generate and adapt joint action–communications policies during mission execution.

b) Computational Tractability: Describe how the proposed approach dynamically bounds the effective per-agent reasoning problem as mission complexity and operational variability increase. TA-1 performers are responsible for compute required for their own training. TA-1 performers should identify the expected compute requirements for their solution for mission-execution inference, including relevant processor, accelerator, memory, storage, software, and runtime dependencies. The Government will work with the TA-2 performer(s) to define the compute resources available for TA-1 solutions during test events. TA-1 performers should design their solutions to execute within the available compute bounds.

c) Communications Efficiency: Describe how the approach determines when to communicate, which agents must communicate, and what information must be exchanged to preserve coordinated action while reducing unnecessary information exchange.

d) Operational Resilience: Describe how the proposed framework will preserve mission performance across diverse mission types and increasing operational variability without overfitting to a single scenario, fixed communications condition, or pre-planned coordination structure.

e) Evidence of Co-Adaptation: Describe the observable policy behavior, instrumentation, and evaluation evidence that will demonstrate adaptation of coordination scope, physical actions, and communications behavior during mission execution. In other words, proposers should describe how their approach dynamically shifts the operating point across computational tractability, communication efficiency, and operational resilience.

f) Integration and Evaluation: Describe how the proposed solution will integrate with the TA-2 training and evaluation environment through the Government-directed interface. Proposers should describe how the solution will support near-real-time and faster-than-real-time repeatable evaluation under common mission conditions. Any abstraction used to enable faster-than-real-time execution must not limit the solution’s ability to execute in real-world operational environments.

In addition to providing the details listed above, in general terms, about their proposed solution, TA- 1 proposals must also leverage the hypothetical mission scenario described in Appendix 1 to describe how their solution will co-adapt autonomy and communication by detailing bullets (a) through (f) discussed above in the context of the mission scenario. Proposers should also discuss how their solution will generalize to other missions beyond the hypothetical scenario in Appendix 1. See Section 3.5 for more information on what content should be included in the proposal.

Proposers should note that approaches that depend on exquisite enabling capabilities, including persistent access to high-bandwidth relays or unconstrained computational resources, are not of interest. Approaches whose primary mechanism depends on periodic and/or intermittent access to high-capacity communications links are also not of interest (e.g., Disruption Tolerant Networking).

Approaches of interest include, but are not limited to: epistemic inference and pragmatic or semantic communications to improve communications efficiency; mission-conditioned decomposition, including factored Dec-POMDP and coordination-graph approaches, state-abstraction methods, including Koopman-based approaches, and hierarchical coordination or learning approaches to improve computational tractability; and multi-agent, hierarchical, federated, or distributed learning approaches that generalize across mission, network, and environmental variability to support operational resilience. Proposals should include a timeline of anticipated technical gains and explain how the proposed approach advances the state of the art in collaborative autonomy under constrained and degraded communications conditions.

1.3.2. Unison Technical Area 2 – Modeling and Simulation Environment Developing and evaluating joint autonomy–communications policies at operationally relevant scale requires exploring diverse missions, combinations of agent tactics and behaviors, and operational variability across repeated trials. TA-2 will develop the common FTRT modeling and simulation environment used to train, stress, and evaluate TA-1 solutions under controlled and repeatable conditions.

TA-2 must provide a scalable, CUI prototype testbed that enables training and evaluation of TA-1 solutions as mission complexity and operational variability increase. The environment must balance five dimensions: scale, faster-than-real-time execution, evaluation-relevant fidelity, repeatability and consistency, and operational variability. These dimensions convey the TA-2 objective: provide a training and evaluation environment that can rapidly exercise TA-1 solutions over the required operating space while preserving the conditions needed for credible, repeatable assessment. TA-2 approaches should:

• Implement a continuous-time event-execution environment to represent physical movement, sensor dynamics, and network conditions for teams of 10 to 200 agents, including air, ground, surface, and subsurface systems as applicable.

• Represent the physical, sensing, and network conditions that materially affect TA-1 autonomy and communications decisions.

• Use a modular architecture and a well-documented API to support black-box integration of TA-1 solutions.

• Achieve an execution factor of at least 100x real-time to enable repeated mission runs, training cycles, parameter sweeps, and stress testing within practical wall-clock time.

• Provide automation to script, configure, execute, and control scenarios, including changes in mission conditions and operational stressors.

• Support replay, regression testing, logging, and data capture needed for diagnosis, scoring, and comparison of TA-1 solutions.

• Produce equivalent computational, communications, and mission behavior across repeated executions under identical initial conditions.

• Model Denied, Degraded, Intermittent, and Limited (DDIL) communications conditions, including bandwidth limits, latency, packet loss, and topology changes.

• Introduce compounding operational variability and uncertainty, including dynamic threats;

expanding operating areas; sensor degradation; blue-force attrition; Position, Navigation, and Timing (PNT) degradation or denial; and adversarial adaptation.

• Support progressive scaling from simple, de-conflicted tasks to tightly coupled, interdependent missions requiring complex team coordination.

• Support the addition of new sensors, terrain, communications, and operational-variability models over the life of the program.

TA-2 approaches should apply fidelity selectively. The testbed must faithfully represent the factors that materially affect mission outcomes, coordination requirements, and communications effects, while allowing abstraction where additional physical realism would increase runtime without materially changing TA-1 evaluation conclusions. TA-2 proposals should describe their simulation architecture and fidelity choices and explain how the approach balances fidelity, scale, FTRT execution, and repeatability. Error! Reference source not found. describes the desired layers of the testbed and their expected fidelity levels.

Layer What it represents Expected Fidelity Physical Layer Propagation, range, interference, attenuation, and channel availability Abstracted: Sufficient to determine whether a physical link is feasible and its resulting characteristics; waveform-level fidelity is not required.

Link / Network Layer

Connectivity and topology dynamics High: Must represent factors that directly affect information availability and coordination Communications Constraint Layer

Enforced bandwidth limits, packet loss, and latency per link

High: Must faithfully enforce the communications bounds experienced by the autonomy solution

Environment / Perception Layer

Agent-observable environment, field of view, localization, and environment state

High: Observations presented to autonomy must reflect operational conditions

Sensor / Data Layer

Visual, thermal, electromagnetic, acoustic, tracks, metadata, etc. that generate perception inputs

Medium: Raw physics may be abstracted through representative sensor products.

Autonomy Interface / API Layer

Standardized state, observations, commands, and messages exposed to the TA-1 solutions

High: Must faithfully reproduce the operational interface

Autonomy Solution Layer

TA-1 solutions, Government- Furnished software baseline, or other black-box autonomy solutions

Native: Must execute the actual software being evaluated

Orchestration Layer

Scenario initialization, faults, uncertainty injection, clocks, synchronization, logging, repeated runs

High: Must support controlled, repeatable experiments and consistent simulation time

Table 1: TA-2 Simulation Layers and Expected Fidelity

The TA-2 testbed will execute mission scenarios that scale along three compounding axes (shown in Error! Reference source not found.): increasing mission complexity, increasing operational variability and uncertainty, and decreasing adaptation windows. The Government team will work with the TA-2 performers to develop dynamic, multi-domain mission sets across representative mission types, including (but not necessarily limited to) civilian search and rescue, combat search and rescue, and intelligence, surveillance, and reconnaissance–indications and warnings. Scenarios will include relevant environmental, communications, and operational stressors, such as terrain blockages and/or communication degradation.

In coordination with the Government-led IPT, TA-1 and TA-2 performers will work to define and document a standardized black-box interface between TA-1 autonomy solutions and the TA-2 testbed. TA-2 will not require access to TA-1 source code, internal models, learned weights, or decision logic beyond that API. The API should support the exchange of relevant state, observations, commands, communications actions, and mission outputs required for execution, training, and evaluation. The TA-2 solution should also include an orchestration layer that configures and initializes scenarios (based on the layers in Error! Reference source not found.), injects operational variability (e.g., sensor faults), manages scenario replay and repeated runs, and captures outputs required for scoring, diagnosis, and reporting.

Additionally, the TA-2 performer may work with the Government to help define, create and/or evaluate an adaptive counter-autonomy solution at the CUI level for Phase 2 execution.

Mission Scenario Axes

Description

Mission Complexity

• Objective: The testbed architecture must support a range of mission types, team scales, and coordination demands without overfitting to a single scenario or operating condition.

• Missions will increase in team scale, coordination demand, and joint-task interdependence.

• At lower complexity, a small number of agents may execute largely de-conflicted or parallel tasks, such as pre-coordinated search sectors in a search-and-rescue mission.

• At higher complexity, teams of tens to hundreds of agents may execute tightly coupled, specialized tasks requiring continuous coordination, such as identification and tracking behaviors to maintain custody of a dynamic target.

• A hypothetical example of a medium complexity mission is detailed in Appendix 1.

Operational Variability

• Objective: The testbed must introduce deviations from planning-time assumptions and enable assessment of whether TA-1 solutions adapt effectively to those deviations at execution-time.

• Scenarios will introduce increasing environmental complexity and uncertainty.

• Lower-variability missions may occur in mapped, static environments with reliable sensing, limited adversarial adaptation, and uninterrupted PNT.

• Higher-variability missions may include unmapped operating areas, expanding mission geometry, dynamic threats, active jamming, blue-force attrition, sensor inaccuracies, PNT disruption, and adaptive adversary behavior.

Adaptation Window

• Objective: The testbed must progressively reduce opportunities for centralized, between-run learning and re-planning to assess whether TA-1 solutions can shift toward execution-time reasoning, coordination, and adaptation.

• Early test events may provide repeated FTRT mission runs with opportunities for centralized data aggregation, learning, and re-planning between clusters of runs.

• Later test events will decrease the frequency and duration of the centralized adaptation cycles, increasingly requiring TA-1 solutions to move learning, reasoning, coordination, and adaptation into mission execution.

Table 2: Mission Scenario Scalability Axes

TA-2 performers will be evaluated using the TA-2 metrics in Section 0. Proposers should describe their innovative approach, addressing:

a) Simulation Architecture: The underlying software architecture, event-execution approach, simulation layers, and modularity.

b) Scale, Speed, and Fidelity: The approach to representing required mission complexity while meeting FTRT performance and preserving evaluation-relevant fidelity.

c) Scenario and Evaluation Framework: The approach to scenario generation, operational-variability injection, repeatability, regression testing, and controlled comparison of TA-1 solutions.

d) API and Interface Design: The proposed black-box integration approach, API design, orchestration responsibilities, and plan for coordination with TA-1 performers and the Government team.

e) Communications and Domain Modeling: The approach to representing multi-domain assets, sensing, DDIL conditions, and the physical and network constraints relevant to TA-1 decision-making.

f) Compute Resources: The amount of compute available for both hosting the underlying testbed as well as executing the TA-1 solutions, including the approach to hosting, scheduling, isolation, and scaling of black-box TA-1 software. The proposed architecture may include cloud-based high-performance computing, on-premises hardware, or another technically justified approach. In coordination with the Government and TA-1 performers, the TA-2 performer should support the establishment of a Government-defined compute profile for TA-1 solution execution during test events.

g) Data Collection and Analysis: Automated data capture, metric calculation, diagnostics, replay, visualization, and reporting required to support Government evaluation.

h) Test Event Support: A plan for testbed availability to TA-1 performers, dry runs, test execution, technical exchanges, and initial and final reporting.

Proposers should note that approaches that satisfy FTRT requirements solely by executing large numbers of real-time simulations in parallel are not of interest. The TA-2 prototype testbed shall provide a controlled fidelity-speed trade space across simulation layers. Approaches of interest include simulation architectures that selectively adjust layer fidelity to achieve the execution speed required for a given training or evaluation objective. These approaches should enable the Government to trade simulation detail for faster execution where reduced fidelity does not materially affect TA-1 autonomy, coordination, communications, or mission-performance conclusions, and to increase fidelity at the expense of reduced execution speed where those effects are material.

1.4. Program Metrics

1.4.1. TA-1 Metrics

The following metrics, summarized in Error! Reference source not found. below, will be used to evaluate TA-1 performer progress during Government test events. Performance will be assessed using scenarios, conditions, baselines, and test procedures established by the Government and provided to performers at least one month prior to each test event. The metrics evaluate computational tractability, communications efficiency, operational resilience, joint co-adaptation, and mission success. Additional performer-proposed metrics may be considered where they provide relevant evidence not otherwise captured by the required measures.

• Computational Tractability (CT)

Reasoning Paths (RP) measure the size of the effective joint decision space that an agent must consider at a single decision step. For a given autonomy solution, RP characterizes the decision-space complexity induced by the world model maintained within the autonomy stack. For agent 𝑖, RP is defined as:

𝑅𝑃𝑖 = ∣𝑆𝑒𝑓𝑓

𝑖 ∣ × ∣𝑂𝑒𝑓𝑓 𝑖 ∣ × ∣𝐴𝑒𝑓𝑓 𝑖 ∣ where ∣𝑆𝑒𝑓𝑓 𝑖 ∣, ∣𝑂𝑒𝑓𝑓 𝑖 ∣, and ∣𝐴𝑒𝑓𝑓 𝑖 ∣ are the effective joint state, observation, and action spaces over which agent 𝑖 must reason. RP captures the combinatorial effect of coordination. As an agent must jointly reason over additional teammates, relevant states, observations, and actions, its effective decision space can grow rapidly. Lower RP values indicate that the agent is reasoning over a smaller effective problem and therefore provides a structural surrogate for computational tractability, provided mission performance is preserved. TA-1 proposers are required to propose mechanisms for measuring/estimating RP for their individual solutions.

Success requires Unison to reduce the effective joint reasoning problem presented to each agent to a mission-specific RP benchmark defined by the Government team. The benchmark will be orders of magnitude smaller than a full-team joint decision space.

The normalized Reasoning Path metric is:

𝑈𝐶𝑇 = max 𝑖 ∈ 𝑁

𝑅𝑃𝑖,Unison

𝑅𝑃Benchmark ≤ 1 where 𝑁 is the set of agents, 𝑅𝑃𝑖,Unison is the RP for agent 𝑖 under a Unison solution, and 𝑅 𝑃Benchmark is the Government-defined benchmark.

• Communication Efficiency (CE)

Bits transmitted measures the amount of explicit inter-agent communication required for an autonomy solution to sustain mission coordination over the mission horizon 𝑇. For a given solution, it is represented by 𝑅(𝑇), the cumulative number of bits transmitted across the team.

Lower 𝑅(𝑇) values indicate that the team is maintaining coordinated mission progress with less explicit information exchange, provided mission objectives are achieved. The metric compares the bits transmitted by a TA-2 solution in a successful mission with that of a Government-provided communication-dependent benchmark autonomy solution executing the same mission.

Success requires Unison to achieve mission objectives while substantially reducing the transmitted bits required for coordination relative to the benchmark. The normalized metric is:

𝑈𝐶𝐸 =

𝑅Unison(𝑇)

𝑅benchmark(𝑇) ≤ 2 × 10―5 where 𝑅Unison(𝑇) is the cumulative number of bits transmitted by a Unison TA-1 solution during a successful mission, and 𝑅benchmark(𝑇) is the cumulative number of bits transmitted by the Government-provided communication-dependent benchmark autonomy solution while executing the same mission.

For reference, using a 50 megabits-per-second application-data baseline consistent with 3rd Generation Partnership Project (3GPP) Unmanned Aerial Vehicle connectivity requirements1, Unison seeks to demonstrate mission success at 1 Kilobit per second or less. This corresponds to a normalized bit rate of no more than 2 × 10―5 of the reference baseline, or a 50,000× reduction in transmitted bits.

• Operational Resilience (OR)

Operational Resilience measures the ability of a team to preserve mission performance as operational variability increases. Mission performance is represented by mission completion time under specified operational variability relative to mission completion time under nominal conditions. Hence, the metric is defined as:

𝑈OR =

𝑡nominal 𝑡uncertainty ≥ 0.8 and 𝑡uncertainty ≤ 𝑇where 𝑡nominal is the mission completion time under nominal conditions, 𝑡uncertainty is the mission completion time under specified operational-variability conditions, and 𝑇 is the required mission horizon or deadline. A value of 𝑈𝑂𝑅 = 1 indicates that operational variability imposed no mission-completion-time penalty relative to nominal conditions. Values below 1 indicate increasing degradation in operational resilience.

The threshold of 𝑈𝑂𝑅 ≥ 0.8 requires successful completion under operational variability within 125 percent of the nominal mission completion time and within the required mission deadline T.

• Joint Co-Adaptation

Joint Co-adaptation measures whether a solution can dynamically balance computational tractability, communications efficiency, and operational resilience as mission complexity and

1 3GPP. Study on Enhanced LTE Support for Aerial Vehicles, 3GPP TR 36.777, Release 15, Sec. 5.1, Table 5.1-1, 2017–2018.

operational variability change, rather than operating at a fixed design point. To demonstrate Joint Co-adaptation, a solution must first meet the applicable threshold for each individual metric:

𝑈𝐶𝑇 ≤ 1, 𝑈𝐶𝐸 ≤ 2 × 10―5, 𝑈𝑂𝑅 ≥ 0.8

In addition, the solution must demonstrate a substantial gain in at least one dimension by achieving one of the following:

𝑈𝐶𝑇 ≤ 0.1, 𝑈𝐶𝐸 ≤ 2 × 10―6, or 𝑈𝑂𝑅 ≥ 0.9

The Joint Co-adaptation metric is:

𝑈joint = min

𝑈𝐶𝑇

0.1

, 𝑈𝐶𝐸

2 × 10―6 , 0.9

𝑈𝑂𝑅

≤ 1 where 𝑈𝐶𝑇, 𝑈𝐶𝐸, and 𝑈𝑂𝑅 are the normalized measures for computational tractability, communications efficiency, and operational resilience, respectively.

This formulation means:

• All three base thresholds must be met.

• At least one dimension must achieve the more demanding stretch threshold.

• The solution must show that it can shift its operating point across the trade space as conditions change.

• Mission Success Rate (MSR)

Mission Success Rate measures the percentage of repeated runs for a given mission in which a solution achieves the Government-defined mission objectives and constraints within the required mission horizon or deadline.

𝑀𝑆𝑅 =

Number of Successful Runs

Total Number of Runs ≥ 90%

The Government team will define the objectives, constraints, and mission horizon applicable to each scenario. Mission Success Rate is the primary effectiveness measure against which reductions in reasoning burden and explicit communications demand are assessed.

• Performer Proposed Metrics

In addition to the above metrics, proposers are encouraged to propose other metrics that might be relevant to their proposed solution. These metrics can capture the unique characteristics of their proposed TA-1 solutions where the above metrics are limited in how they are able to evaluate the solution. The Government team will work with the performer on refining their proposed metric(s) and determining how best to evaluate during Test Events.

TA-1 Metrics Metric Rationale Success Criterion Evaluation Method Threshold Computational Tractability

(UCT)

Bound per-agent reasoning while achieving mission objectives as mission complexity and operational variability

Each agent’s effective decision burden remains computationally tractable as team size, operational variability, and coordination demands

Reasoning Paths as a surrogate for per-agent decision-space complexity, compared against a government-defined benchmark.

Maintain UCT ≤ 1 as mission complexity and operational variability increase increase. grow.

Communication Efficiency

(UCE)

Reduce communication dependence without unacceptable loss of coordination or mission performance.

Teams achieve mission objectives with substantially lower explicit information exchange than a communication-dependent baseline.

Reduced bit transmission, which compares bits transmitted during mission execution to a communication-dependent baseline.

Month 7:

𝑈𝐶𝐸 ≤ 10―3

compared to Gov benchmark

Month 15:

𝑈𝐶𝐸 ≤ 10―4

compared to Gov benchmark

End of Phase 2: 𝑈𝐶𝐸 ≤ 2 × 10―5 compared to Gov benchmark

Operational Resilience

(UOR)

The point of co-adaptation is to sustain mission performance as mission complexity and operational variability increase.

Solutions continue to meet mission objectives under degraded communications and changing mission conditions.

Operational Resilience, a mission-level measure of how closely mission completion under operational variability tracks mission completion under nominal conditions.

Maintain UOR ≥ 0.8 of nominal conditions as mission complexity and operational variability increase

Joint Co-adaptation

Balance between computational tractability, communication efficiency, and operational resilience as mission complexity and operational variability change

Meet threshold on the three metrics above and show measurable adaptive tradeoff across the trade space, under different stress conditions, rather than operate at a fixed design point.

Same evaluation method as described above for each of the three metrics

Per metric threshold AND one of the following:

UCT ≤ 0.1, OR

UOR ≥ 0.9, OR

UCE ≤ 2×10⁻6 by program end (or 10x less than the phased Month UCE threshold)

Mission Success Rate (MSR)

Achieve mission objectives specified at planning time within a required deadline.

Meet mission objectives specified at planning time within a required deadline.

The fraction of runs that satisfy mission objectives and associated constraints

Maintain MSR ≥ 90% as mission complexity and operational variability increase

Other Metrics…