Nature / Nature Communications Class Low Altitude Autonomous Systems Paper Planning v1: From Engineering Systems to Falsifiable Scientific Questions

The goal of this article is not to continue to expand the selection of engineering topics, but to strictly judge: based on the current three articles, which directions have the opportunity to be promoted to the Nature / Nature Communications level.

The three articles I am currently working on are:

Article	Current Properties	Normal Submission Positioning
Paper A: Conflict-free path planning / PPO-MAPPO / Low-altitude conflict resolution	Tactical security control	T-ITS / T-RO / ICRA / IROS
Paper B: Three-layer hierarchical dispatch of hundreds of UAVs	Urban low-altitude transportation system operation	TR-C / T-ITS
Paper C: Fisher Information-Driven UAV 3DGS Active Sensing	Active Sensing and Digital Twins	T-RO / ICRA / IROS / CVPR Directions

Write the conclusion clearly first: **In its current form, A/B/C are not Nature-level papers. ** The key to Nature Communications must be to shift from “engineering systems to do better” to “discovering, verifying and explaining a falsifiable scientific law.”

1. The threshold of Nature / Nature Communications

The Nature journal requires that the article has outstanding scientific importance and can arouse the interest of scientists outside the field [1]. Nature Communications is a multidisciplinary journal that aims to publish high-quality research with important advancements in various fields [2]. This means that for low-altitude UAV papers to enter this level, they must not only meet:

Propose a new scheduler;
Propose a new reinforcement learning collision avoidance method;
Propose a low-altitude large model agent;
A few points higher than baseline in simulation;
Make a system demo or benchmark.

Nature / Nature Communications is more concerned about:| Questions that need to be answered | Common writing methods for engineering papers | Nature-level writing methods | |---|---|---| | Why this question is important | A certain algorithm is better | Solving a scientific bottleneck that is common across systems | | What is the contribution | Propose methods | Discover patterns, prove boundaries, and establish measurement methods | | What’s the evidence | Simulated curves and ablation | Theory + statistical confidence + real world calibration | | Is it falsifiable | Not easy | Must be reproduced or overturned by other data, cities, and hardware | | Scope of influence | UAV/ITS/Robotics community | Common concerns of transportation science, complex systems, safety certification, and embodied intelligence |

For us, the core judgments are:

A/B/C/G/I/J/K are mostly engineering papers themselves; only when they serve scientific issues such as rare-event safety measurement, capacity phase transition, energy-intelligence scaling law, they may enter the field of Nature Communications.

2. Relevant high-level precedents

2.1 Precedent for autonomous driving safety verification

The NADE work of Shuo Feng et al. was published in Nature Communications, proposing to use naturalistic and adversarial environments to test autonomous driving intelligence [3]. The follow-up Dense Reinforcement Learning for Safety Validation was published in the main issue of Nature. The core is not “training a stronger agent”, but turning autonomous vehicle safety validation into accelerated rare-event measurement, and claims to achieve 10^3 to 10^5 times acceleration without losing unbiasedness [4].

The direct inspiration from this is: **The most promising opportunity for Nature Communications in the low-altitude UAV direction is not the planner itself, but the calibrable safety verification method. **Nature Communications also published related work on “curse of rarity”, which clarified that the rarity of rare safety-critical events in high-dimensional space will hinder deep model learning and verification [5]. This is consistent with core challenges for low-altitude UAVs: a combination of collision, near-miss, non-cooperating vehicle intrusion, communication failure, and wind disturbance are all low-probability but high-consequence events.

2.2 Precedent of swarm robots and complex systems

Nature Communications has also recently published articles in the direction of swarm robotics/collective intelligence. For example, snail-inspired robotic swarms demonstrate the adaptability of groups in outdoor unstructured environments [6]; the collective intelligence model for swarm robotics applications puts swarm robotics into a more general collective intelligence modeling framework [7].

This shows that Nature Communications does not exclude robotic systems, but what it values is not “the robot system can run”, but the collective adaptation, scaling, emergence, phase transition, and universal behavior behind the system.

2.3 Low-altitude large model and embodied intelligent background

A review of large models of low-altitude economy has broken down low-altitude systems into facility networks, information networks, route networks and service networks, and pointed out that large models need to be combined with edge computing, communication networks and trusted autonomous systems [8]. SINGER, FlightGPT, UAV-VLN, OpenVLA, RT-2 and other work descriptions aerial embodied intelligence, VLA/VLN, robot foundation model are developing rapidly [9] [10] [11] [12] [13].

But these efforts also bring risks: **LLM/VLA/Agent direction is overheated, and simply doing LowAltitudeGPT or CloudBrain-Agent can easily be considered engineering packaging. ** According to Nature Communications, the model must become an “instrument for measuring scientific laws” rather than the protagonist of the paper.

---## 3. Three independent Claude review conclusions

In this round, three independent Claude review agents were invited to strictly review from the three perspectives of Nature editor, complex transportation/safety science, and embodied AI/robotics/edge intelligence. The consensus of the three reviewers is as follows.

Review Perspective	Most Approved Direction	Clear Rejection Direction	Key Reasons
Nature / Nat Commun Editor’s Perspective	B, C; D only holds true when rare-event estimates can be certified	A, G, I, J, K	Engineering performance improvements are not Nature-level contributions and must have universal laws or measurement methods
Complex system / traffic safety perspective	D first, B second, C third	G, I, J, K, unreconstructed A	D has rare-event estimator precedent; B has capacity phase change potential
Embodied AI / edge intelligence perspective	B+K fusion, scaling-law version of D, I	Separate G/J/K, normal I	I/J/K defaults to engineering; only energy-intelligence or embodied scaling law may upgrade

Strict consensus from three reviewers:

**The official issue of Nature is currently unrealistic. **
**Nature Communications does not have no chance at all, but the problem definition must be changed. **
**D: Low-altitude safety-critical rare-event accelerated testing is the strongest candidate. **
**B: The low-altitude traffic capacity phase transition/scaling law is the second candidate. **
**B+K: Low-altitude embodied cluster energy-intelligence scaling law is a high-risk, high-yield candidate. **
**A, G, I, J, and K are independent engineering papers and should not conflict with Nature Communications. **
**C Only when the cross-scenario active sensing information-cost theoretical boundary is proposed and verified by real UAV field, will there be a boundary opportunity. **

4. The most recommended main line: Certifiable rare event safety measurement in low-altitude autonomous airspace

4.1 Suggested topics

Certifiable Rare-Event Safety Measurement for Autonomous Low-Altitude Airspace

Chinese can be written as:Certifiable Rare Safety Event Measurement for Autonomous Low Altitude Airspace

4.2 Core scientific issues

Safety incidents with low-altitude autonomous systems are rare events. In the real world, catastrophic events rarely occur, but when they occur, the consequences are serious. Directly estimating the accident rate using Monte Carlo or ordinary simulation will encounter a curse of rarity: most samples are normal flights, and the truly valuable safety-critical samples are overwhelmed by the massive number of normal samples [5].

The core question is:

Is it possible to construct an accelerated safety verification method suitable for low-altitude UAV multi-agent systems, which can significantly reduce the sample size while still providing risk estimates that are calibrated, with confidence intervals, and can be verified by real data?

This is not “generating hazardous scenarios” but measuring the probability of low-altitude autonomous system failure.

4.3 Target contribution

Contribution	Required Form
Low-altitude rare-event space definition	Non-cooperative UAV, communication degradation, positioning error, wind disturbance, corridor conflict, vertiport near loss, emergency insertion
Accelerated sampling theory	importance sampling / rare-event density / adversarial but naturalistic distribution
Estimator guarantee	unbiasedness or bounded bias; variance reduction; confidence interval
Simulated versus real calibration	Simulated failure distribution aligned with real/physical near-loss events
Certifiable output	failure rate, LoWC/NMAC risk, scenario criticality, policy-specific safety envelope

4.4 Relationship with existing Paper A/B/C| Already published articles | Role in Nature Communications main line |

|---|---| | Paper A | Tested conflict resolver / safety controller, not a main contributor | | Paper B | Provides low-altitude traffic flow, queue, and density conditions for generating system-level risk states | | Paper C | Provides perception uncertainty, map missingness, and 3DGS scene errors for constructing perception-induced risk | | Paper D/F | Become the core of the main paper: scenario coverage + accelerated rare-event safety validation | | Paper G/I/J/K | Only as an optional intelligent system under test or engineering support, not as a Nature-level main contribution |

4.5 Data source

Nature Communications level cannot rely solely on self-study simulation. It is recommended to use three layers of data:

Data layer	Source	Function
Public safety proxy	NASA ASRS database containing voluntary safety reports by aviation frontline personnel and UAS crews [14]
Public UAS reports	FAA UAS sighting reports / FOIA electronic reading room [15]
Air traffic baseline	OpenSky Network ADS-B / Mode S crowdsourced air traffic data [16]
Urban environment	OpenStreetMap / VGI urban road network, buildings, POI and semantic functional areas [17] [18]
Controlled physical data	Indoor/outdoor multi-UAV testbed, injecting non-cooperative UAV, communication delay, positioning noise
Simulation exposure	Self-developed low-altitude corridor/world generator, expanded to 10^7-10^8 equivalent exposure samples	Let’s be honest here: ASRS, FAA UAS sighting, and OpenSky are not perfect low-altitude UAV fleet data. Their purpose is to calibrate risk type, spatial distribution, and near-loss event statistical priors. The true system-level failure rate still needs to be supplemented by simulation and hardware-in-the-loop.

4.6 Experimental design

The main experiment should not be written as “Our method is safer” but rather as “Can we reliably measure safety risks?”

Experiment	Question	Success Criteria
Brute-force Monte Carlo comparison	Whether the speedup estimate is unbiased or calibrable	In small-scale brute-force enumerable scenarios, the estimate falls within the Monte Carlo confidence interval
Acceleration multiplier experiment	Is rare-event curse really alleviated	Under the same error, the sample size can be reduced by more than 10^3 level
Variance reduction experiment	Is the estimator stable	The CI is narrower under multiple seeds, and the variance reduction is significant
Tested across algorithms	Whether applicable to A/RRT/ORCA/CBF/MAPPO	Not dependent on a single planner
Cross-city topology	Whether to generalize across cities	Maintain calibration in OSM-derived multi-city topology
Hardware-in-the-loop	Whether there is a realistic anchor point	The sequencing of real multi-machine/controlled near-loss events is consistent with the simulation criticality
Counterfactual verification	Whether the found dangerous scenario is real is key	After modifying variables such as communication/wind/path, the risk changes are in line with the model predictions

4.7 Evaluation indicators- accident/collision probability estimate;

Loss of Well Clear rate;
Near Mid-Air Collision proxy;
variance reduction ratio;
acceleration factor;
calibration error;
effective sample size;
confidence interval width;
scenario naturalness;
failure-mode coverage;
sim-to-real rank correlation.

4.8 Fatal Risks

Risk	Severity	Mitigation
No real data anchor	Fatal	Get ASRS/FAA/OpenSky proxy first + self-built physical testbed
Biased accelerated sampling	Fatal	Small-scale brute-force calibration + theoretical estimator correction
Hazardous scenes are not naturalistic	High	Constrained sampling distribution with real reports and urban structure
Only prove that a planner is unsafe	High	Evaluate at least 5 types of planners/policies
The paper is regarded as a simulation benchmark	Fatal	The main line must be risk measurement, not benchmark ranking

5. Second candidate: Low-altitude traffic capacity phase transition and congestion collapse law

5.1 Suggested topics

Capacity Phase Transitions in Autonomous Low-Altitude Traffic Networks

Chinese can be written as:

Capacity phase transition and congestion collapse rules in autonomous low-altitude transportation networks

5.2 Core scientific issues

Paper B is currently a three-tier scheduling system. For Nature Communications, we must switch from “scheduler” to “complex system law”:> Is there a tipping point in low-altitude transportation networks from free flow to congestion collapse? This critical point is determined by what variables are demand intensity, vertiport capacity, charging capacity, corridor separation, and communication reliability? Is there a scaling law that is reproducible across city topologies?

This is similar to the traffic fundamental diagram of ground transportation, but the object becomes a three-dimensional low-altitude corridor + vertiport + charging + fleet scheduling.

5.3 Possible scientific discoveries

Scientific proposition	Form requiring verification
There is a low-altitude traffic critical load	When demand/capacity exceeds the threshold, backlog, delay, and LoWC risk increase nonlinearly
The bottleneck leadership mechanism is switchable	Low load is dominated by demand, medium load is dominated by charging, and high load is dominated by corridor/vertiport
Three-layer scheduling changes the phase transition point	H-LyraUAV is not just better, but expands the stable area
Multimodal fallback changes critical behavior	UAV-ground transfer turns abrupt collapse into smoother degradation
Critical index of urban topology impact	Grid, radial, strip, and high-speed corridor cities have different capacity scaling

5.4 Minimum threshold for experimentation- 10 / 20 / 50 / 100 / 200 / 500 / 1000 UAV continuous scale scanning;

5 / 10 / 20 / 50 vertiports;
low / medium / peak / shock demand;
Multi-city OSM topology;
charging capacity, corridor separation, communication degradation scanning;
At least one real OD proxy: NYC TLC, Chicago taxi, logistics order proxy, emergency incident proxy;
Report phase diagram instead of single performance table.

5.5 Why this might work Nat Commun

It’s no longer “We propose H-LyraUAV” but:

We found that there are predictable capacity boundaries and phase transition laws for the stable operation of autonomous low-altitude transportation systems, and provided an interpretable queuing stability theory and cross-city empirical verification.

If this pattern can be replicated in different cities, different scheduling strategies, and different traffic modes, there is a possibility for Nature Communications.

5.6 Fatal Risks

The biggest risk is that the result degenerates into the known conclusion of classic queuing / network flow.
If it is just “the greater the load, the greater the delay”, there is nothing new. It must be proven that low-altitude systems produce critical behaviors that are not fully described in ground transportation or traditional queuing networks due to the coupling of 3D separation, charging, vertiport, multimodal transfer, and communication degradation.

6. High-Risk Third Candidate: Energy-Intelligence Co-scaling Law of Low-altitude Embodied Clusters

6.1 Suggested topics

Energy-Intelligence Scaling Laws in Embodied Low-Altitude Swarms

Chinese can be written as:

Energy-intelligent co-scaling law in low-altitude embodied clusters

6.2 Why not ordinary K / I / JK alone is for inference acceleration, and the default is system engineering.

The separate I is aerial VLA/VLN, the default is Robot/AI engineering.
The J alone is the LowAltitudeGPT nudge, which defaults to Domain Model Engineering.

But if you combine them into scientific questions, there is a possibility:

Is there a reproducible Pareto front or scaling law between mission success rate, collective coordination quality, communication overhead, inference latency and energy consumption in low-altitude multi-UAV embodied systems?

The Nature Communications opportunities for this path come from “law”, not from “LLM”.

6.3 Measurable variables

Variables	Examples
System scale	Number of UAVs N = 5-500
Intelligent resources	Model size, token budget, planning horizon, tool-call depth
Energy resources	onboard compute power, communication energy, flight energy
Task quality	success, delay, LoWC risk, coverage, mapping quality
Architecture strategy	cloud-only, edge-cloud split, onboard fallback, hybrid agent

6.4 Minimum evidence threshold

At least three types of tasks: conflict resolution, emergency dispatch, and active perception;
At least three types of deployment: cloud GPU, edge workstation, onboard Jetson/embedded GPU;
At least four model scales: small / 8B / 14B / 32B / API teacher;
Draw energy-intelligence Pareto frontier across tasks;
Give a theoretical explanation: why some split policies are close to the lower bound;
At least small-scale real UAV or hardware-in-the-loop verification.

6.5 Strict review and judgment

This is a high-risk direction. It is not currently possible to start writing Nature Communications directly. It is suitable as strategic direction 12-24 months out, provided we first have:1. Paper A/B/C/D/G tool chain; 2. Recordable cloud-brain workload; 3. Real or semi-real UAV hardware platform; 4. Complete energy consumption and delay measurement.

7. It is not recommended to go in the direction of Nature Communications alone.

Directions	Strict judgment	A more suitable place
Paper A: PPO/MAPPO Conflict-Free Planning	Engineering Algorithms Unless Risk Measurement of D is Incorporated	T-ITS / T-RO / ICRA / IROS
Paper C: FIM-3DGS active sensing	Strong method paper, but Nature level requires cross-domain information theory laws	T-RO / ICRA / IROS / CVPR related
Paper G: CloudBrain-Agent	System integration, easily considered agent hype	AAAI / IJCAI / T-ITS
Paper I: Aerial VLA/VLN	The common version is Embodied Navigation Engineering	CoRL / RSS / ICRA / IROS
Paper J: LowAltitudeGPT	Vertical fine-tuning, not a scientific discovery	T-ITS / Applied Intelligence / AAAI workshop
Paper K: Inference acceleration	System optimization, not low-altitude science per se	MLSys / SenSys / TMC / IoT Journal

It’s not that these directions aren’t worth pursuing, but rather that Nature Communications should not be directly targeted. They should serve as supporting tools or counterpart engineering papers for the D/B/B+K main line.

8. Recommended execution route

8.1 Immediate execution: Nature Communications pre-research package

Priority is given to D: low-altitude rare-event safety measurement.

Weeks 1-2:- Collate ASRS / FAA UAS sighting / OpenSky / OSM data availability;

Define low-altitude rare-event taxonomy; -Extract failure modes from existing 76 million exploration logs;
Confirm whether it can be used as a small-scale physical multi-machine testbed.

Weeks 3-6:

Implement brute-force Monte Carlo small-scale ground truth;
Implement importance sampling / adversarial naturalistic sampling;
Prove the bias / variance / confidence interval of the estimator;
Make first-round risk estimates for A*/RRT*/ORCA/CBF/MAPPO.

Weeks 7-10:

Cross-city OSM topology replication experiment;
Add communication degradation, wind, positioning error;
Distribution alignment with FAA/ASRS event taxonomy;
Do hardware-in-loop or small-scale practical flights.

Weeks 11-16:

Develop a Nature Communications style narrative:
- Rare-event safety is the bottleneck in the deployment of low-altitude autonomous systems;
- This article proposes a calibrable acceleration measurement method;
- Ability to estimate multi-UAV failure risk with 10^3 level sample efficiency;
- Methods remain calibrated across planners, cities, and disturbance types;
- The results provide inspiration for low-altitude safety certification.

8.2 Synchronization preparation: capacity phase change version of B

Paper B continues to advance according to TR-C, but the experiment requires an extra set of Nature Communications data:

Do continuous density scans;
Record non-linear changes in backlog / delay / risk;
Draw phase diagram;
Analyze stable/unstable boundary;
try to extract scaling exponent;
Compare critical behavior of UAV-only, ground-only, multimodal fallback.### 8.3 Things not recommended
Do not package CloudBrain-Agent as Nature level;
Don’t write “low-altitude AGI” as the core selling point;
Don’t commit to training a low-altitude foundation model from scratch;
Don’t just use simulation to claim certifiable safety;
Do not package 3DGS active sensing as Nature Communications alone, unless the theoretical boundaries and real external fields are complemented.

9. Nature Communications version of the paper skeleton

It is recommended to write a preliminary draft with D as the main line.

9.1 Abstract

Problem: The deployment of low-altitude autonomous systems is limited by rare safety-critical events that are difficult to observe and verify.
Methods: A calibrable rare-event accelerated safety measurement framework is proposed.
Results: Estimate collision / LoWC / near-miss risk with significantly fewer samples under multiple low-altitude airspace, planner, and disturbance conditions.
Significance: Provide reproducible measurement methods for autonomous low-altitude airspace safety certification.

9.2 Introduction

Narrative chain:

The low-altitude economy relies on the safe operation of a large number of UAVs;
Safety-critical events are rare and direct testing costs are unacceptable;
There is a precedent for accelerated validation in autonomous driving, but low-altitude airspace is higher-dimensional and more unstructured;
Existing UAV simulation/planning/scheduling efforts lack calibrable risk measurements;
This article proposes low-altitude rare-event safety measurement and verifies its statistical validity.

9.3 Methods- Low-altitude event space;

Naturalistic prior;
Criticality function;
Accelerated sampling distribution;
Risk estimator;
Confidence interval;
Sim-to-real calibration; -Planner/policy under test.

9.4 Results

Figure	Content
Fig. 1	General diagram of low-altitude rare-event safety measurement framework
Fig. 2	rare-event taxonomy corresponds to real reported/simulated events
Fig. 3	Small-scale brute-force calibration: estimates vs Monte Carlo ground truth
Fig. 4	Speedup factor and variance reduction
Fig. 5	Cross-planner risk measurement
Fig. 6	Cross-city/perturbation generalization
Fig. 7	Hardware-in-the-loop or true near-missing ordering verification

9.5 Discussion

Must discuss:

The method cannot replace real regulatory certification, but it can significantly improve the efficiency of pre-certification testing;
There is a deviation between the simulation distribution and the real world;
ASRS/FAA data is voluntary reporting and there is selection bias;
Real low-altitude operation logs are needed in the future;
Significance for low-altitude supervision, scene library, and UAV planner benchmarking.

10. Final judgment

Currently, the most worthy investments in Nature Communications’ pre-research are:

**Main line D: Certifiable low-altitude rare-event safety measurement. **
**Subline B: Low-altitude traffic capacity phase transition/congestion collapse law. **
** Long-term high risk B+K: low-altitude embodied cluster energy-intelligence scaling law. **It is currently not recommended to contact Nature Communications directly:

A’s PPO/MAPPO conflict resolver;
Common FIM-3DGS NBV of C;
CloudBrain-Agent for G;
Common aerial VLA/VLN of I;
J’s LowAltitudeGPT;
Normal inference speedup by K.

In one sentence:

Nature Communications-level papers cannot be written as “We made a stronger low-altitude UAV system”, but must be written as “We discovered and verified a falsifiable law between the safety, capacity, or energy consumption intelligence of low-altitude autonomous systems.”

11. References

[1] Nature. Editorial criteria and processes. URL: https://www.nature.com/nature/for-authors/editorial-criteria-and-processes

[2] Nature Communications. Aims & Scope. URL: https://www.nature.com/ncomms/aims

[3] Shuo Feng, Xintao Yan, Haowei Sun, Yiheng Feng, and Henry X. Liu. “Intelligent Driving Intelligence Test for Autonomous Vehicles with Naturalistic and Adversarial Environment.” Nature Communications, 12:748, 2021. DOI: 10.1038/s41467-021-21007-8. URL: https://www.nature.com/articles/s41467-021-21007-8[4] Shuo Feng et al. “Dense Reinforcement Learning for Safety Validation of Autonomous Vehicles.” Nature, 615:620-627, 2023. DOI: 10.1038/s41586-023-05732-2. URL: https://www.nature.com/articles/s41586-023-05732-2

[5] Shuo Feng et al. “Curse of Rarity for Autonomous Vehicles.” Nature Communications, 15:4808, 2024. DOI: 10.1038/s41467-024-49194-0. URL: https://www.nature.com/articles/s41467-024-49194-0

[6] Da Zhao, Haobo Luo, Yuxiao Tu, Chongxi Meng, and Tin Lun Lam et al. “Snail-Inspired Robotic Swarms: A Hybrid Connector Drives Collective Adaptation in Unstructured Outdoor Environments.” Nature Communications, 15:3647, 2024. DOI: 10.1038/s41467-024-47788-2. URL: https://www.nature.com/articles/s41467-024-47788-2[7] Alessandro Nitti, Marco D. de Tullio, Ivan Federico, and Giuseppe Carbone et al. “A Collective Intelligence Model for Swarm Robotics Applications.” Nature Communications, 16:6572, 2025. DOI: 10.1038/s41467-025-61985-7. URL: https://www.nature.com/articles/s41467-025-61985-7

[8] Jinpeng Hu, Wei Wang, Yuxiao Liu, and Jing Zhang. “Large Model in Low-Altitude Economy: Applications and Challenges.” Big Data and Cognitive Computing, 10(1):33, 2026. DOI: 10.3390/bdcc10010033. URL: https://www.mdpi.com/2504-2289/10/1/33

[9] Maximilian Adang, JunEn Low, Ola Shorinwa, and Mac Schwager. “SINGER: An Onboard Generalist Vision-Language Navigation Policy for Drones.” arXiv:2509.18610, 2025. URL: https://arxiv.org/abs/2509.18610[10] Hengxing Cai et al. “FlightGPT: Towards Generalizable and Interpretable UAV Vision-and-Language Navigation with Vision-Language Models.” EMNLP, 2025. DOI: 10.18653/v1/2025.emnlp-main.338. URL: https://aclanthology.org/2025.emnlp-main.338/

[11] Pranav Saxena, Nishant Raghuvanshi, and Neena Goveas. “UAV-VLN: End-to-End Vision Language guided Navigation for UAVs.” arXiv:2504.21432, 2025. URL: https://arxiv.org/abs/2504.21432

[12] Moo Jin Kim et al. “OpenVLA: An Open-Source Vision-Language-Action Model.” arXiv:2406.09246, 2024. URL: https://arxiv.org/abs/2406.09246

[13] Anthony Brohan et al. “RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control.” arXiv:2307.15818, 2023. URL: https://arxiv.org/abs/2307.15818

[14] NASA. Aviation Safety Reporting System Database Online. URL: https://asrsdbol.arc.nasa.gov/[15] Federal Aviation Administration. Unmanned Aerial System (UAS) & Small Unmanned Aerial System (sUAS) FOIA Electronic Reading Room. URL: https://www.faa.gov/foia/electronic_reading_room/uas

[16] Martin Strohmeier, Xavier Olive, Jannis Lübbe, Matthias Schäfer, and Vincent Lenders. “Crowdsourced Air Traffic Data from the OpenSky Network 2019-20.” Earth System Science Data, 2021. URL: https://essd.copernicus.org/articles/13/357/2021/

[17] Michael F. Goodchild. “Citizens as Sensors: The World of Volunteered Geography.” GeoJournal, 69:211-221, 2007. DOI: 10.1007/s10708-007-9111-y.

[18] Geoff Boeing. “OSMnx: New Methods for Acquiring, Constructing, Analyzing, and Visualizing Complex Street Networks.” Computers, Environment and Urban Systems, 65:126-139, 2017. DOI: 10.1016/j.compenvurbsys.2017.05.004.

12. Appendix: This execution plan1. Don’t change the current submission route of A/B/C, and continue to make them solid according to TR-C/T-ITS/T-RO/ICRA.

Use Paper D as the main line of Nature Communications pre-research and rename it to low-altitude rare-event safety measurement.
Simultaneously transform the experiment of Paper B to output capacity phase diagram and congestion collapse evidence.
Postpone writing the Nature version of LowAltitudeGPT / CloudBrain-Agent / AerialVLA separately, and only use them as the intelligent system under test or data generation tool.
Complete data availability audit within two weeks: ASRS, FAA UAS sightings, OpenSky, OSM, existing 76 million exploration logs, available hardware platforms.
Complete a small-scale proof-of-concept within four weeks: brute-force Monte Carlo vs accelerated estimator.
If the estimator cannot be calibrated, immediately downgrade it to the T-ITS/T-RO safety test paper and do not continue to submit it to Nature Communications.