Key Takeaways

• Autonomous operations (Level 0) should account for at least 70% of all missions — this is the fundamental principle of MUM-T scalability
• Four control levels span from device autonomy through close-range LOS, squad NLOS, to battalion remote supervision, each with distinct link requirements and operator roles
• The Authority Token Model provides a clean FSM for control handoff: IDLE → AUTONOMOUS → CLAIM_REQUEST → CONTROL_GRANTED → SHARED_ASSIST → PREEMPTED → RECOVER
• Video streaming must use a perception-aware 3-tier approach (Control Preview + Focus Stream + Forensic Stream) with semantic ROI — naive full-frame high-bitrate transmission is unsustainable in contested networks
• A Network-Adaptive Ladder dynamically adjusts encoding from 1080p 30fps down to 240p+metadata based on RTT, packet loss, jitter, and queue occupancy
• The Single Control / Multi Observe model prevents control conflicts while enabling collaborative annotation across observers

1. Control Mode Hierarchy

Four Control Scenarios

• Level 0: Autonomous — Unmanned Systems Operate Independently

Examples:

• UAV patrol
• Robot dog inspection
• UGV path tracking

Human only sends: go to waypoint A

The device completes the loop autonomously — this is the default state.

This must account for 70%+ of operations.

Otherwise, no network can sustain the load.

This is the first principle:

Minimize continuous manual control.

• Level 1: Close-Range LOS Companion Control

Typical: Team member directly controls robot dog / small UAV.

Range: 10~100m

Link: StarFlash / WiFi Direct / 600M Mesh

Characteristics: Low latency, high interactivity.

Suitable for:

• Fine manipulation
• Doorway reconnaissance
• Tight-space traversal

• Level 2: Squad NLOS Control

Typical: Squad controls indoor robots via Mesh.

Range: 100m~2km

Dependency: 600M Mesh

Characteristics: Limited bandwidth

Therefore: Intelligent video scheduling is essential

• Level 3: Battalion Remote Control

Cross-squad / Cross-region

Dependency: 5G / SAT / relay mesh

Suitable for:监督控制，不是细粒度 teleop。

This is critical.

Semi-closed loop

Operator: Provides speed/attitude suggestions

Device’s local controller executes the closed loop.

Example:

Not:

left stick → motor PWM
But rather:
"move 0.5 m/s left"


2. Control Architecture: Authority Token ModelEach device maintains a Control FSM
IDLE
↓
AUTONOMOUS
↓
CLAIM_REQUEST
↓
CONTROL_GRANTED
↓
SHARED_ASSIST
↓
PREEMPTED
↓
RECOVER



Control Priority:



Controller
Priority




Safety Control (Automatic)
P0


Squad Leader
P1


Authorized Team Member
P2


Battalion Remote
P3（Default）



Preemption Mechanism
Example:
Battalion Requests takeover:
Flow:
Request
→ Squad Approve
→ Token Migrate
→ State Sync
→ Control Handoff
No direct seizure.
Except: Emergency Override



3. Internal Autonomy Control (Operator Disconnection)





Mode 1: Hot-Standby Operator


Robot maintains:
Primary
Secondary
Primary掉线 >500ms：
Automatic switchover.


Mode 2: Autonomous Fallback


If no one takes over:
Enter: Safe autonomy
Examples:
UAV → Hover / Return to relay point
Robot Dog → Hold position + Local scan
UGV → Stop + Watchdog



Mode 3: Adjacent Takeover


Nearest node automatically receives:
“Take Control Offer”



4. Video Links: The Real Bottleneck
Control requires video assistance, but naive full-frame high-bitrate streaming won’t work.
Must use:
Perception-aware Video Transport
This is the core strategy.


Recommended: Three-Tier Video Streaming





Layer A: Control Preview


Low bitrate, always on
Parameters:

480p
10fps
300~800kbps

Provides situational continuity.




Layer B: Focus Stream


ROI high-quality region, dynamically triggered
Examples:
Doorway
Target
Operation area
Bitrate: 2~6 Mbps




Layer C: Forensic Stream


High-quality recording
Local storage
On-demand upload

5. Optimal ROI Strategy

Semantic ROI.
That is:
Perception module generates:
{
  "bbox": [x,y,w,h],
  "class": "door",
  "priority": 0.92
}
Encoder follows the ROI map.


Applicable to:
1. Manual control area — auto-enhance the region the joystick points to


2. 检测Target-人 / 门 / 热源 / 障碍


3. Operation point — robotic arm grasp location



6. Dynamic Encoding Strategy
Network-Adaptive Ladder


Priority状态
Green: Ample bandwidth
1080p @ 30 fps


Yellow: Light congestion
720p @ 20 fps


Orange: Severe congestion
ROI only


Red: Control keep-alive
240p + metadata


Switching based on:
RTT
Packet loss
Jitter
Queue occupancy
Not just bandwidth.

7. Multi-Operator Observation & Control Design

Single Control / Multi Observe
That is:一个人控，多人看


Example:
UAV-3
Controller: Squad-2-A
Observers:
- SquadLeader
- CommandHQ
- Squad-2-B
Observers: Can view, can annotate, cannot send control commands.


Feature: Collaborative Pointer

8. Control Link Recommendations




Close-Range LOS


Primary: StarFlash / WiFi Direct
Secondary: 600M
Encoding: Low-latency I-frame sparse




Squad NLOS


Primary: 600M Mesh
Secondary: 5G
Encoding: ROI + adaptive GOP




Battalion Remote Control


Primary: 5G
Secondary: SAT / 1.4G
Control: High-level commands, not low-level teleoperation

Building MUM-T systems for defense or industrial applications? Aomway provides custom UAV, UGV, and communication solutions optimized for multi-domain teaming. Contact us at [email protected] for expert consultation.
9. Control Software Roadmap



1. Control Authority Manager
Token management
Authorization
Audit logging


2. Stream Adaptation Engine
Dynamic bitrate
ROI
Multi-stream switching


3. Handoff Manager
Control handoff
State synchronization


4. Policy Engine
Rules:
uav:
  remote_direct_control: false
  squad_override: true

自治优先 + 监督控制 + TokenAuthorization + ROI感知视频 + 分层控制域

Devices execute autonomously; humans intervene only at critical moments, with intervention rights strictly scheduled.




Have questions about this article? Feel free to contact us at [email protected] — we’re happy to help!