Let’s begin with an industry signal.

Anthropic, a company founded by former core executives from the ChatGPT team and focused on AI programming tools and large-model applications, is becoming one of the most closely watched companies in the global AI wave. Behind its rapid growth is a very clear signal: the value of AI in programming is no longer limited to “helping write a few lines of code”; it is driving structural changes in productivity.

Efficiency gains of 10x, or even 100x, are already happening in reality.

For the drone and robotics R&D industry, the value of AI has long gone beyond programming assistance. It is penetrating key stages such as requirement breakdown, solution design, UI design, structural design, algorithm development, testing, and debugging, accelerating the entire R&D workflow.

This is also the direction we, as a UAV solution provider, continue to explore and practice. At present, we have integrated AI tools such as Figma, GPT, Cursor, and Codex into the R&D process for core modules, functional modules, and complete systems.

01

Early-Stage R&D

Clearer Requirement Breakdown and More Efficient Solution Design

The starting point of R&D is requirement breakdown and solution planning.

This stage is where issues such as vague requirements, unclear solutions, and high communication costs are most likely to arise. Once there is a misunderstanding in the early stage, it can easily lead to repeated revisions or even full rework later.

The core value of introducing AI at this stage is to quickly align requirements, generate feasible solutions, reduce decision-making costs, and remove obstacles for subsequent R&D work.

Requirement Breakdown and Transformation

We feed customers’ drone or robot requirements directly into large AI models, such as endurance requirements, payload capacity, and flight accuracy for industrial inspection drones, or motion control and environmental perception requirements for service robots.

By combining this with our long-term accumulated industry requirement database, AI can complete requirement breakdown in a short time and convert originally vague customer statements into specific technical indicators.

For example, when a customer says they want to “improve drone inspection efficiency,” AI can further break this down into actionable indicators such as flight speed, endurance, image recognition accuracy, obstacle-avoidance response speed, and data transmission stability.

At the same time, AI can automatically match industry standards and past successful cases, identify unreasonable aspects of the requirements, and provide optimization suggestions, reducing later-stage R&D rework from the source.

Preliminary Overall Solution Design

Based on the broken-down technical indicators, AI can quickly generate an overall architecture plan for a drone or robot, including hardware selection direction, software architecture planning, and control-logic design.

For example, in a robot motion-control solution, our architects first provide a code architecture diagram, program flowchart, input/output interface documents, and coding standards. AI then combines these with existing technical assets to generate a preliminary control-algorithm framework.

In the early stage of R&D, architecture design and process design are the most critical tasks. Only when the direction is correct can subsequent development efficiency truly improve.

Laying the Foundation for AI-Generated Code

The control-algorithm architecture diagrams, rule highlights, and solution highlights generated in the early stage are not one-off documents.

They are added to the subsequent AI project engineering files as important references for AI to understand the project background, code boundaries, and development standards. In this way, when AI generates code later, the output can be more accurate, more stable, and more aligned with engineering implementation requirements.

02

Core Design

AI Penetrates Key R&D Stages

Software UI: Accelerating the Journey from Prototype to Code

The UI of a drone ground-control terminal and the UI of a robot interaction interface directly affect the user experience.

Traditional UI design requires designers to manually build prototypes, adjust layouts, annotate specifications, and then hand them over to front-end developers. The whole process is time-consuming and often leads to issues such as insufficient design fidelity and inconsistent adaptation.

With Figma’s AI plugin ecosystem, we further accelerate the UI design workflow.

• Prototype Generation

Designers only need to enter UI design requirements, such as: “A drone ground-control terminal that includes four major modules: flight-parameter display, route planning, real-time image transmission, and fault alarms; the style should be clean and professional, and it should support both desktop and tablet devices.”

Figma’s AI plugins can quickly generate multi-page prototypes, including interface layouts, button styles, font combinations, and other elements, saving the time required to build the basic structure from scratch.

• Optimization and Iteration

AI can also identify issues in UI design, such as cluttered layouts, inconsistent button spacing, unclear module hierarchy, and poor adaptation results, and then provide optimization suggestions.

Designers only need to provide revision requests, such as “adjust the layout of the flight-parameter display module and enlarge the font size of key data,” and AI can quickly complete the adjustment.

With this workflow, the efficiency of organizing a single-page design can be improved by more than 30%.

• Code Conversion

Through third-party AI plugins for Figma, such as Builder and Anima, completed UI prototypes can be directly converted into front-end code.

This greatly reduces the handoff cost between UI design and front-end development, while also reducing issues related to poor interface reproduction.

Especially in the development of web interfaces for drone ground-control terminals, the delivery cycle from prototype to code can be compressed from about one week to roughly half a day.

Structural Design: Enabling Solutions to Take Shape Faster

The airframe structure of a drone and the body structure of a robot directly affect product stability, endurance, and payload capacity.

Traditional structural design requires engineers to manually create models, analyze mechanical performance, and repeatedly verify solutions. It places high demands on experience and professional expertise.

By leveraging GPT-series models and image-generation capabilities, we introduce AI into the early stage of structural design to help engineers complete concept development and solution communication more quickly.

• Generation of Structural Concept Drawings and Block Diagrams

In the early design stage, engineers only need to enter structural requirements, such as: “Drone airframe structure, lightweight design, carbon-fiber material, compatible with XX-model core module, and 20% endurance improvement.”

AI can then quickly generate structural block diagrams and concept drawings, presenting information such as airframe layout, module mounting positions, and stress-point distribution.

These drawings can be used for internal reviews, technical communication, and customer solution presentations, reducing the time engineers spend manually drawing and repeatedly explaining solutions.

• 3D Modeling Assistance

GPT can also combine our structural design specifications to provide modeling ideas for engineers.

For example, it can generate modeling parameters for drone wings and airframe shells, annotate key dimensions and assembly relationships, and assist in optimizing wing curvature and the airframe’s center of gravity to improve flight stability and reduce drag.

At the same time, by incorporating multiphysics coupling analysis, AI can help the team identify potential issues in structural design earlier and reduce later-stage validation costs.

It should be noted that AI-generated structural diagrams are mainly used for solution communication and design reference; they cannot be used directly for engineering production. Final precision modeling still needs to be completed by engineers using professional EDA tools.

At this stage, AI’s core role is to reduce the cost of solution communication, improve early-stage design efficiency, and help the team discover issues earlier.

Algorithms and Programming: Truly Connecting AI to Drone Development

Programming and algorithm development are the most central and challenging parts of drone and robotics R&D.

This is also where AI programming tools can demonstrate their greatest value. Tools such as Cursor and Codex are not merely code-completion plugins; they are intelligent programming assistants that can participate in environment setup, code generation, testing, debugging, and issue fixing.

Through tools such as Cursor and Codex, we connect AI to development computers and target onboard computers, covering the complete workflow from development environment setup to code debugging.

• AI Remotely Accesses the Onboard Computer

To enable AI to truly participate in drone development, the hardware and network environment must first be connected.

Specifically, the onboard computer on the drone and the development computer are placed on the same local area network, and this local network must have Internet access. With a complete hardware network topology in place, AI can access the onboard computer through SSH remote login.

Once the SSH login user has sufficient permissions, AI can read, understand, modify, and debug both the low-level code and upper-layer application code on the target onboard computer.

The prerequisite is that the overall architecture design is clear and the boundary conditions are well defined. The better the architecture is built, the higher the quality of the development tasks completed by AI.

• Development Environment Setup

Traditional development environment setup requires engineers to manually install dependencies, configure compilers, and set up debugging interface environments. This is not only time-consuming, but also prone to version inconsistencies and dependency conflicts.

Codex can use natural-language instructions and SSH remote login to the target system to automatically complete development environment configuration.

For example, an engineer can enter: “Set up a development environment for drone flight-control algorithms, support the C language, install the XX compiler and XX debugging tool, and configure the onboard computer communication interface.”

Codex can then complete the relevant configuration and automatically handle some dependency issues, reducing manual intervention.

In multi-person collaborative projects, this can effectively ensure environment consistency and reduce collaboration costs.

• Code Auto-Generation

Automatic code generation is the core advantage of AI programming and a key driver of R&D efficiency.

Based on the specific R&D requirements of drones and robots, we input functional requirements into Cursor or Codex, and the corresponding code can be generated quickly.

However, this does not mean that programmers are being replaced. On the contrary, programmers must first complete the overall code architecture design and clarify module boundaries, interface specifications, and development direction. Only when the boundaries are clear can AI-generated code be reliable.

(1) Onboard Computer Code

AI can participate in code development for the target onboard computer, generating code related to data acquisition, signal transmission, command execution, and more.

For example, for perception modules such as cameras and radar, it can generate data acquisition and parsing code to enable real-time transmission of information such as images and distance. It can also generate robot motion-control code to implement functions such as gait control and steering control.

(2) Auxiliary Function Code

AI can also quickly generate auxiliary modules such as log output, data storage, and fault detection.

At the same time, it is also well suited for batch modifications and code refactoring. For example, when adapting old-version code to a new core module, Codex can automatically identify adaptation points and complete the modifications, reducing a large amount of repetitive work.

Unlike traditional code-completion tools, Codex can understand the entire code repository and independently complete more complex programming tasks. It can not only write code, but also run it, debug it, locate issues, and propose optimization suggestions.

This is equivalent to equipping every R&D engineer with an “AI programming assistant,” freeing engineers from basic repetitive work and allowing them to focus more energy on core algorithm optimization and product innovation.

• Code Testing and Debugging

AI can not only generate code, but also participate in automated testing and debugging.

Cursor and Codex can automatically generate unit-test code covering core modules such as flight-control algorithms and onboard programs, while simulating multiple operating scenarios to detect bugs, logic vulnerabilities, and performance issues.

When code abnormalities occur, AI can also assist in locating the problem and providing repair suggestions. For example, when a “signal delay” issue appears in flight-control code, AI can analyze it from perspectives such as communication interface parameters, task scheduling logic, and data-processing links, then propose optimization plans.

In addition, AI can participate in code reviews, checking for performance issues, security vulnerabilities, and coding-standard problems, thereby reducing deployment risks in advance. It can also automatically generate API documentation to support team collaboration and later maintenance.

03

AI Taking Over R&D Brings More Than Efficiency Gains

Based on industry trends and our own practice, the value of applying AI to drone and robotics R&D is mainly reflected in three aspects.

Efficiency Improvement

R&D cycles that originally required months may be compressed to weeks or even less. The efficiency gains already proven by AI in programming are now extending across the full R&D process for drones and robots.

Cost Reduction

Through AI-assisted design, automated code generation, virtual simulation, and automated testing, repetitive manual input can be reduced, lowering the overall cost of R&D, testing, and mass production.

Quality Improvement

AI’s capabilities in automatic inspection, continuous optimization, and standardization can reduce human error and improve product stability and consistency.

In the future, we will continue to deepen the application of AI throughout the entire R&D workflow.

On one hand, we will deeply integrate AI with robots’ autonomous learning and drones’ intelligent inspection capabilities to develop more intelligent products, such as industrial drones with autonomous obstacle avoidance and decision-making capabilities, as well as robots with environmental adaptability.

On the other hand, we will continue to optimize the application of AI programming tools such as Cursor and Codex in drone R&D, driving core algorithm iteration and R&D process automation.

AI is changing the way drones and robots are developed.

It is not merely an auxiliary tool; it is gradually becoming part of the R&D system across stages such as requirements, design, programming, testing, and debugging.

AI is reshaping the underlying logic of drone and robotics R&D. As a UAV solution provider, we will continue to integrate AI tools such as Figma, GPT, Cursor/Codex, and others into the full R&D workflow, making development more efficient, products more intelligent, and implementation more reliable. In the future, we will continue to deepen AI application innovation, build drone and robot products with stronger core competitiveness, and drive the industry toward a new stage of intelligent R&D.