On October 26, 2026, Taiwan's Ocean Affairs Council (OAC) announced that a research team from the Department of Aeronautics and Astronautics at National Cheng Kung University, which the council supports, had completed a simulation in which 20 drones flew simultaneously in two swarms. The simulation used three types of fixed-wing unmanned aircraft capable of vertical takeoff and landing (VTOL). Overseas media have described the technology as a way to counter Chinese gray-zone pressure. However, the OAC says the main intended user is not the military but the Coast Guard Administration (CGA), the maritime law enforcement agency under its authority. The OAC calls the result an important milestone in moving from controlling a single aircraft to coordinating multiple swarms, but 20 aircraft were not actually flown. The announcement also does not say when the next stages, hardware-in-the-loop (HIL) testing and at-sea trials, will take place or how much funding they require.
The research is drawing attention against the backdrop of increasingly active Chinese government vessels around Taiwan. In July 2026, reports said that the number of Chinese coast guard ships, survey vessels and similar craft detected around the main island of Taiwan reached a record 244. In this environment, technology that coordinates multiple drones to monitor wide sea areas is seen as a way to extend the CGA's surveillance capability. However, operating 20 simulated aircraft at sea will require several more stages of verification.
What did the 20-drone, three-type simulation verify?
The formal title of the research, translated from Chinese, is roughly "AI mission planning for swarm flight of multi-type, multi-fleet VTOL fixed-wing drones for three-dimensional coast guard operations." It aims to develop AI that assigns missions to multiple swarms made up of different aircraft types and flies them autonomously, extending CGA maritime surveillance into the air. According to the OAC, the study integrates AI, VTOL fixed-wing drones, flight dynamics and swarm coordination control. It targets three types: the JUMP 20, the V-BAT and the Flexrotor. All take off and land vertically and fly as fixed-wing aircraft once airborne.
The team first recreated the exterior of each of the three types as a 3D model and analyzed its aerodynamic parameters. It then ran six-degree-of-freedom (6-DOF) flight simulations to build flight-characteristic and control models for each type. Six degrees of freedom means movement in three directions (forward-back, left-right, up-down) plus three kinds of rotation (nose up and down, rolling left and right, and heading). Modeling these allowed the team to calculate how each aircraft responds to control commands.
Coordinating multiple drones requires an understanding of each one's flight characteristics. VTOL fixed-wing aircraft in particular change attitude and aerodynamic behavior significantly when moving from hovering to level flight, and different types respond differently to the same command. The V-BAT, for example, is a tail-sitter that takes off standing vertically and changes attitude in the air to transition to level flight. Reflecting these differences between types in the control models is a prerequisite for operating different aircraft together.
For coordinated control, the team adopted multi-agent reinforcement learning (MARL). In this approach, each aircraft is treated as a learning agent that optimizes task allocation and flight paths while taking into account the behavior of nearby aircraft and changes in the environment. The team combined it with leader-follower formation control, in which other aircraft adjust their positions relative to a lead aircraft, and with control for avoiding moving obstacles. The OAC says the swarm can change its flight path autonomously even when it encounters sudden environmental changes or obstacles.
The achievement was flying a total of 20 aircraft, divided into two swarms, simultaneously in a simulated environment. The OAC describes it as an important step from single-aircraft control to multi-swarm coordination and says it overcame challenges in autonomous coordination among different aircraft types. To prevent collisions, the control algorithm incorporates three-dimensional distance between aircraft and a minimum vertical separation of 100 feet (about 30 m). Some English-language media have described 100 feet as the minimum distance between aircraft in three-dimensional space, but the OAC announcement explicitly specifies only vertical separation. In any case, what has been confirmed is a computational result, not a successful simultaneous flight of 20 real aircraft.
Next steps are HIL and at-sea testing, with timing and budget undisclosed
The OAC lists hardware-in-the-loop (HIL) testing, prototype integration and flight tests in actual sea areas as the next development stages. Through these, it plans to verify flight reliability under varying sea conditions and the potential for practical use. However, the October 26 announcement gives no specific information on when each test will begin, which sea areas will be used, or the size of the subsidy. The details may appear in other materials such as budget documents or research plans, but this announcement alone does not show when at-sea testing will start.
HIL testing connects real hardware, such as the flight control computer and sensors, to a simulation environment and checks its operation under virtual flight conditions. Software-only simulation can also reproduce communication delays and processing times, but those values depend on the models and assumptions set. Incorporating real hardware makes it possible to confirm computation time, signal transmission delay and processing limits as measured values. It is an important stage for checking whether a control algorithm that worked in software will also work with the computing power and communication conditions of real aircraft.
In drone swarm control in particular, each aircraft must track the positions and movements of its neighbors and correct its path in a short time. If communication or computation delays grow, the swarm may fail to maintain spacing or may be slow to decide on obstacle avoidance. HIL testing can examine such problems without the risk of a real crash or collision. Moving from simulation to HIL testing, then to prototype verification and flight tests in the real environment, is a standard way of reducing development risk step by step.
The OAC also acknowledges that maritime operations pose particular challenges. The CGA's surveillance area is vast and sea conditions change readily, and it has traditionally relied heavily on personnel and patrol vessels. Whether different aircraft types can fly as simulated and hold formation over the sea, affected by wind and waves, is a question for future at-sea trials. This announcement is only a progress report on research the OAC subsidized; it does not give a deployment schedule or procurement scale.
CGA drone acquisition and Chinese vessel activity in 2026
In 2026, Taiwan's CGA drone acquisition and its response to Chinese government vessels have advanced in parallel. On June 16, OAC Minister Kuan Bi-ling said the CGA would spend NT$2.1 billion over several years to build up four types of drone systems: short-range, medium-range, ship-based and special-mission. NOWnews reported the remarks, which are aimed at strengthening the CGA's maritime surveillance capability.
Several developments followed in July. The South China Morning Post (SCMP) reported that the CGA had completed inspection and delivery of 45 training multicopters and was deploying them to units so personnel could learn basic piloting skills. That same month, Shield AI announced that the V-BAT had been selected in the competitive selection for the CGA's ship-based drone program. Meanwhile, activity by Chinese coast guard ships, survey vessels and similar craft around Taiwan also increased. The Taipei Times reported on October 19, citing Taiwanese government data, that the number of vessels detected around the main island in July reached a record 244.
| Period | Chinese coast guard ships, survey vessels and similar craft detected around Taiwan's main island |
|---|---|
| July 2025 | 58 |
| May 2026 | 30 |
| June 2026 | 55 |
| July 2026 | 244 |
Source: Taipei Times (October 19, 2026). Warships are not included in the count.
The July 2026 figure of 244 was a sharp rise from 30 in May and 55 in June, and it far exceeds the 58 recorded in July 2025. According to the Taipei Times, coast guard ships made up about 60% of the vessels detected. However, the figure counts Chinese coast guard ships, survey vessels and similar craft detected around Taiwan's main island and does not include warships. It should not be read as a measure of China's overall military activity.
In October, the CGA continued to receive new equipment and respond in the field. On October 7, a V-BAT that had arrived at Taipei Port was shown to Minister Kuan and National Security Council Secretary-General Joseph Wu. Shield AI formally announced the first delivery on October 13 and named Aerospace Industrial Development Corporation (AIDC) as its Taiwanese partner. The October 7 showing date comes from reports that reproduced Shield AI's announcement. On October 18, the CGA also announced that it had driven away a total of 16 Chinese coast guard ships, survey vessels and fishing boats that had been operating simultaneously off the east, south and southwest of Taiwan's main island and around Pratas (Dongsha).
Nineteen days after the CGA showed off its first V-BAT on October 7, the OAC announced the completion of the 20-drone coordinated flight simulation. Both relate to strengthening the CGA's drone operating capability, but the OAC's announcement of the research results does not mention the V-BAT delivery. It is therefore unclear whether the simulation is directly linked to the V-BAT acquisition plan.
CGA's 12 aircraft versus the navy's 280: different V-BAT procurement plans
According to the SCMP, the CGA has budgeted NT$1.2 billion (about US$39 million by the paper's conversion) to acquire 12 V-BATs and six command-and-control systems. The first delivered aircraft will be used mainly for training, and the rest are scheduled for delivery over the coming months. The V-BAT is a ducted-fan tail-sitter VTOL drone developed by US company Shield AI. It can take off vertically from the limited space on a ship and transition to level flight. The company says it can fly for more than 12 hours.
Taiwan's navy, meanwhile, is pursuing a separate, much larger procurement. According to the Ministry of National Defense's fiscal 2027 report as relayed by the Taipei Times, the navy plans to spend NT$36 billion over four years under the "Yuan Xiao" (遠鴞) program to procure 280 V-BATs, 140 control systems and 82 transport vehicles. The CGA and the navy plan to adopt the same V-BAT, but the operating entities and procurement scales differ greatly.
| Entity | Number of aircraft | Accompanying equipment | Budget (NT$) | Current stage |
|---|---|---|---|---|
| CGA (law enforcement) | 12 | 6 command-and-control systems | NT$1.2 billion | Budgeted. First aircraft received; remainder due within months (SCMP) |
| Navy | 280 | 140 control systems, 82 transport vehicles | NT$36 billion (four years) | Procurement plan in the Ministry of National Defense's fiscal 2027 report (Taipei Times) |
The navy plans 280 aircraft against the CGA's 12, a difference of about 23.3 times in numbers. However, because the budgets include different accompanying equipment, one cannot simply divide budget by aircraft count to compare per-unit prices. The navy's 280 aircraft are also still at the procurement-plan stage. Note, too, that all of these figures come from media reports rather than the original budget documents.
On the software that controls autonomous flight, separate development is under way on the military side. TaiwanPlus reported on August 7 that Taiwan's National Chung-Shan Institute of Science and Technology (NCSIST) had installed Shield AI's autonomous flight software "Hivemind" on its domestically built "Mighty Hornet III" drone and completed a fully autonomous search mission. In the test, three aircraft each flew their assigned areas and landed while maintaining contact with one another. That was a flight test with real aircraft, whereas the 20-aircraft coordinated flight at National Cheng Kung University remains a simulation. However, because the aircraft configurations, test conditions and purposes differ, one cannot judge technical superiority from aircraft counts or test stages alone.
There is also a link between the CGA and NCSIST in maritime unmanned systems. TechNews, citing a Nikkei Asia report based on an anonymous senior Taiwanese official, says the CGA plans to procure 25 unmanned surface vessels and two unmanned underwater vehicles from NCSIST. NCSIST is working with US firm MARTAC and aims to test the first vessels in the first half of 2027. The V-BAT has also been adopted in Japan: Shield AI announced in January 2025 that the Japan Maritime Self-Defense Force had selected it as a ship-based drone.
Thus in Taiwan, several programs are advancing in parallel: CGA drone acquisition, the navy's large-scale procurement and NCSIST's autonomous flight tests. The National Cheng Kung University research falls among them as an effort the OAC supports with CGA operations in mind. Rather than lumping everything together as "Taiwan's drone swarm development," it is necessary to distinguish each program's operating entity, purpose and technical maturity.
What remains before a 20-drone swarm can operate at sea
Behind the CGA's push to expand its drone surveillance is the need to respond to the activities of Chinese coast guard ships. In an interview published by NOWnews in April, a CGA deputy director-general said Chinese coast guard ships enter prohibited and restricted waters around Kinmen 46 times a year, each time with four ships. Recently, after entering, they have been splitting into two groups of two ships heading in different directions, dispersing CGA patrol vessels and wearing them down. For a CGA with limited vessels and personnel to watch wide sea areas, targets that split in several directions are a heavy burden.
The OAC expects three capabilities from this swarm flight technology: coordination of multiple aircraft, wide-area surveillance and continuous surveillance. The concept is to use drones as "eyes in the sky" for CGA personnel, extending monitoring to areas that patrol vessels and coastal surveillance equipment alone find hard to cover. The SCMP also points out that introducing drones could supplement the monitoring range of vessels and coastal sensors. However, the OAC announcement does not describe how drone swarms would specifically be used against the tactic of Chinese coast guard ships splitting up.
Preparations for real operations have begun on both the equipment and technology sides. In the National Cheng Kung University study, a coordinated flight simulation of 20 aircraft combining three different types has been completed. The CGA has received 45 training multicopters, and the first V-BAT has arrived. The remaining V-BATs are due over the next few months. But this does not mean these assets have reached the stage where a 20-aircraft swarm can operate at sea.
One point to watch is the specific schedule and budget for HIL and at-sea testing. Another is how many real aircraft of the three studied types can be obtained. First delivery of the V-BAT has been confirmed, and the JUMP 20 was reportedly provided free of charge by the United States in 2023 and used by the CGA for operator training. For the Flexrotor, however, no information confirms that the CGA has received real aircraft. It is therefore unclear at present whether a setup can be established that combines the three simulated types in actual coordinated flight at sea.
Training of the personnel who will operate them is also essential. Minister Kuan has stressed that introducing advanced equipment is not enough, and that it must be accompanied by improved skills and training for personnel. The 45 multicopters delivered in July are meant to build basic piloting skills, and Shield AI has said it will train CGA pilots and maintenance staff for V-BAT operations. Moving from piloting a single aircraft to coordinated multi-aircraft operations will require, in addition to such training, procedures for assigning missions, sharing information between aircraft and responding to anomalies.
To connect this research result with actual equipment deployment, important conditions include the OAC disclosing the schedule and budget for the next test stage, V-BAT deliveries progressing, and personnel acquiring the skills to operate multiple aircraft. This, however, is an organization based on the published research content and equipment status, not conditions the OAC has formally set for at-sea testing. The October 26 announcement also does not include a specific test plan for flying 20 drones simultaneously in reality.
The significance of the result is that it verified, in simulation, a basic technology for using AI to coordinate drones with different flight characteristics. Challenges remain for practical use, however, including stable flight at sea, the reliability of communication and control, securing aircraft and training operating personnel. If the timing and scale of HIL and at-sea testing become clear, it will be possible to judge more concretely whether a 20-drone swarm can be used in actual maritime surveillance.