# US 2026/0065792 A1 · System and method for strategic airspace deconfliction using cooperative multi-agent reinforcement learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0065792 A1
- Document ID: US20260065792A1
- Status: Published application
- Kind: A1
- Application number: 19/383,639
- Filed: 2025-11-08
- Published: 2026-03-05
- Earliest claimed: 2025-11-08
- Inventors: Richard Joseph Mitchell
- Web page: https://patentgrep.com/patent/US20260065792A1

## CPC

- G08G 5/80: Anti-collision systems
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/092: Reinforcement learning
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G08G 5/22: located on the ground
- G08G 5/26: Transmission of traffic-related information between aircraft and ground stations
- G08G 5/55: for a single aircraft
- G08G 5/57: for unmanned aircraft
- G08G: TRAFFIC CONTROL SYSTEMS (guiding railway traffic, ensuring the safety of railway traffic B61L; arrangement of road signs or traffic signals E01F9/00; radar or analogous systems, sonar systems, lidar systems specially adapted for traffic control G01S13/91, G01S15/88, G01S17/88; {radar or analogous systems, sonar systems, lidar systems specially adapted for anti-collision purposes G01S13/93, G01S15/93, G01S17/93})
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system and method for strategic airspace deconfliction is disclosed, centered on a Cooperative Multi-Agent Platform (CMAP) embedded within an Automated Data Service Provider (ADSP) operating within a federated Unmanned Aircraft Systems (UAS) Traffic Management (UTM) network. The system's inventive feature resides in a specific technical architecture that synergistically integrates a real-time safety constraint into a strategic negotiation engine. A Regulatory Compliance Monitor (RCM) generates a real-time Conflict Risk Score that is incorporated directly into the Local Observation Vector of a Multi-Agent Reinforcement Learning (MARL) policy network. The MARL engine is thereby configured to select a strategic negotiation primitive (e.g., BID, YIELD, TRADE) that is dynamically determined based on this real-time Conflict Risk Score. This transforms abstract resource allocation into a safety-aware, risk-adaptive protocol, allowing an ADSP agent to maximize fleet efficiency while guaranteeing collective safe separation

## Claims

### Claim 1

A method for safety-adaptive strategic deconfliction in a federated Unmanned Aircraft Systems Traffic Management (UTM) network, the method comprising the steps of:
a. receiving, by a processor of an Automated Data Service Provider (ADSP) agent from an associated Regulatory Compliance Monitor (RCM), a real-time Conflict Risk Score quantifying a risk of non-conformance for a fleet of Unmanned Aircraft Systems (UAS);
b. generating, by the processor, a Local Observation Vector, wherein said Local Observation Vector comprises said real-time Conflict Risk Score and a historical negotiation context;
c. inputting, by the processor, said Local Observation Vector into a trained Multi-Agent Reinforcement Learning (MARL) policy network;
d. selecting, by the MARL policy network, a strategic negotiation primitive from a predefined action space, wherein said action space comprises at least a ‘BID’ primitive and a ‘YIELD’ primitive, and wherein said selection is dynamically determined based at least in part on said real-time Conflict Risk Score; and
e. transmitting, by an Intent Negotiation Module (INM), the selected strategic negotiation primitive to a peer ADSP agent via a USS Network Application Programming Interface (API) to dynamically resolve a potential conflict.

### Claim 2 (depends on 1)

The method of claim 1, wherein the predefined action space further comprises a ‘PROPOSE TRADE’ primitive, a ‘PROPOSE MODIFICATION’ primitive, and an ‘ACCEPT/REJECT’ primitive.

### Claim 3 (depends on 1)

The method of claim 1, wherein the MARL policy network was trained using a composite reward function, said composite reward function comprising:
a. a negative safety component (Rsafety) applied as a penalty when said Conflict Risk Score exceeds a predefined threshold;
b. a positive efficiency component (REfficiency) based on minimizing fleet flight time; and
c. a positive negotiation component (RNegotiation) based on successful outcomes of said negotiation primitives.

### Claim 4

A Cooperative Multi-Agent Platform (CMAP) system for safety-adaptive strategic deconfliction, configured for deployment within an Automated Data Service Provider (ADSP) infrastructure, the system comprising:
a. a Regulatory Compliance Monitor (RCM) module configured to:
i. perform real-time conformance monitoring for a fleet of Unmanned Aircraft Systems (UAS), and
ii. calculate a real-time Conflict Risk Score based on said conformance monitoring;
b. a processor; and
c. a non-transitory memory storing a trained Multi-Agent Reinforcement Learning (MARL) policy network and executable instructions that, when executed by the processor, configure the processor to function as a MARL Engine, said MARL Engine configured to:
i. receive said real-time Conflict Risk Score from the RCM;
ii. generate a Local Observation Vector comprising said real-time Conflict Risk Score and a historical negotiation context;
iii. select a strategic negotiation primitive from a predefined action space by inputting said Local Observation Vector into said trained MARL policy network, wherein the selected primitive is dynamically determined based at least in part on said real-time Conflict Risk Score; and
d. an Intent Negotiation Module (INM) configured to transmit the selected strategic negotiation primitive to a peer ADSP agent.

### Claim 5 (depends on 4)

The system of claim 4, wherein the MARL policy network was trained utilizing a Centralized Training, Decentralized Execution (CTDE) architecture and a value decomposition algorithm, said algorithm being one of a Value Decomposition Network (VDN) or a QMIX algorithm.

### Claim 6

A non-transitory computer-readable medium storing executable instructions that, when executed by a processor of an Automated Data Service Provider (ADSP) operating in a Unmanned Aircraft Systems Traffic Management (UTM) network, cause the processor to perform the steps of:
a. receiving, from an associated Regulatory Compliance Monitor (RCM), a real-time Conflict Risk Score quantifying a risk of non-conformance;
b. generating a Local Observation Vector, wherein said Local Observation Vector comprises said real-time Conflict Risk Score and a historical negotiation context;
c. inputting said Local Observation Vector into a trained Multi-Agent Reinforcement Learning (MARL) policy network;
d. selecting, by the MARL policy network, a strategic negotiation primitive from a predefined action space, wherein said selection is dynamically determined based at least in part on said real-time Conflict Risk Score; and
e. transmitting the selected strategic negotiation primitive to a peer ADSP agent via an Intent Negotiation Module (INM).
