PFN and ACSL Reach Stage for Autonomous Flight-Route Generation Demo Under NEDO K Program

Preferred Networks (PFN) and ACSL announced on September 10, 2026, that they had reached a stage where they can demonstrate AI-based automatic flight-route generation under the New Energy and Industrial Technology Development Organization’s (NEDO) Economic Security Important Technology Nurturing Program (K Program) project, “Research and Development of Autonomous and Distributed Control Technologies for Small Unmanned Aircraft.” The companies have been conducting research and development since October 2025 and are working on the project’s second phase, which runs from fiscal 2025 through fiscal 2027. The announcement outlined an approach for generating proposed flight plans based on mission requirements and the surrounding environment.

The announcement by PFN and ACSL is significant because it points to using AI to replace some of the route-design work in drone operations that has relied heavily on people. According to the announcement, the technology is intended for use in situations requiring rapid decisions and efficient operations, including disasters, emergencies and infrastructure inspections. The work is also proceeding under NEDO’s K Program, a public initiative, and combines flight-plan generation and reconstruction, object recognition and obstacle avoidance. This means the question is whether the technology can be implemented across operations as a whole, rather than as a standalone function.

The new development in this announcement is that PFN and ACSL have advanced their research on autonomous and distributed control technologies for small unmanned aircraft to the point where they can demonstrate automatic flight-route generation. However, the work remains at the research and development stage; it is not a completed operational system or the launch of a mass-produced aircraft. It is more accurate to characterize the achievement as automation of a preflight process—creating flight plans—rather than as a move to having AI directly fly the aircraft. In a related article, [Japan’s METI and NEDO report selection of 13 companies and groups for GENIAC support for robot foundation model development](/en/news/meti-and-nedo-report-selection-of-13-companies-and-groups-for-geniac-r-e92f5f), data collection and structuring were the focus in the context of robot foundation models. The PFN-ACSL project also presents a structure in which mission requirements, flight conditions and the surrounding environment are used to generate proposed routes, suggesting that discussion of foundation models is beginning to connect with flight planning for real aircraft. However, unlike the tactile sensing and task data discussed for ground robots under the broader heading of “autonomous control,” this project puts greater emphasis on replanning in response to environmental changes and obstacle avoidance. From an industry-structure perspective, the issue here is less the aircraft itself than the chain of controls: mission description, environmental perception, flight-plan generation, replanning and coordination among multiple aircraft. If some of the route design that operators have set individually is automated, value will depend not only on aircraft performance but also on how many conditional scenarios AI can handle and how much environmental change it can accommodate through replanning. ACSL’s focus on developing and mass-producing domestically made industrial drones, combined with PFN’s focus on vertical integration that includes AI foundation models, makes for a clear development setup connecting hardware and software. But actual differentiation will depend on how broadly the technology can be applied to field conditions. Questions that remain include how reliably automatic route generation will work across different numbers of conditions and types of terrain, which missions will be prioritized after implementation in real aircraft, and how far the scope of control will extend when integrated with multi-aircraft coordination. Public information also does not yet specify the demo’s evaluation metrics, the timing of testing in real-world environments or the extent to which the technology will be incorporated into mass-produced aircraft.

In the project, PFN and ACSL are developing an initial aircraft as well as technologies including AI-based flight-plan generation, replanning in response to environmental changes, object recognition and obstacle avoidance. The small unmanned aircraft are intended for both peacetime and emergency situations, with missions in six categories: situational surveys, inspections, security, search, surveying and communications. The demo is a technical verification at the research and development stage. The companies say they will next work on implementing the technology in small unmanned aircraft, testing it in environments designed to reflect real-world conditions, and integrating it with control technologies designed for coordination among multiple aircraft. ACSL describes itself as a developer and mass producer of domestically made industrial drones, while PFN says it vertically integrates the AI technology value chain.

ACSL develops and mass-produces domestically made industrial drones, while PFN says it vertically integrates from generative AI foundation models to semiconductors. This setup could create an opportunity in Japan to connect drone airframes, control AI and operational data within the same framework. At the same time, a key challenge will be how far the companies can advance implementation in aircraft and testing in real-world environments.