Rapid human-like AI brain production: Robot crafts robot intelligence at a rate 20 times quicker than human counterparts.
In a groundbreaking development, computer scientist Peter Burke has designed a robot that can train itself using Generative AI. This remarkable project, which took roughly 20 times less time than Burke and his team's previous four-year project, Cloudstation, is redefining what robots can build using AI.
The heart of this innovation is an AI-generated drone software, which leverages generative AI models like ChatGPT and others (e.g., Claude, Gemini) to automatically write all the necessary code for a fully autonomous drone control system. This system, named WebGCS, runs on a Raspberry Pi computer inside the drone, making its control dashboard accessible online.
This project entails two "robots": - The AI models running on cloud or local machines that generate the code. - The drone executing the AI-authored software autonomously.
The approach integrates autonomous capabilities such as real-time mapping, GPS tracking, autonomous mission planning, execution, and safety protocols—all orchestrated via a web interface hosted onboard. Notably, no human wrote any of the final flight control code, meaning the drone’s "brain" was created by AI itself.
The implications for autonomous systems are significant. This method produces code approximately 20 times faster than traditional human coding practices, signifying a substantial acceleration of robotics development. It also marks a fundamental paradigm shift towards robots co-designing and building other robots’ cognitive systems, increasing autonomy and flexibility.
Potential expansion into various fields demanding autonomous aerial operations such as spatial AI and aerial imaging is another promising aspect. However, these advances raise important safety, ethical, and operational concerns due to the ability of drones to self-generate control logic and operate in unpredictable environments. Manual overrides remain necessary but highlight challenges in trust and reliability.
The work points towards generalizable autonomy beyond specific task robotics, hinting at future robots capable of broader adaptive intelligence. Burke, who compares his project to Arnold Schwarzenegger's character in the movie "The Terminator", hopes to prevent a condition like the Terminator from occurring in the future.
When flying, the drone hosts its control dashboard, which can be accessed online. The lower brain of the drone is its firmware, which manages flight. The intermediate brain (GCS) is responsible for live maps, mission planning, and drone settings. The higher brain helps the drones avoid obstacles on their own.
The fourth sprint led to the successful generation of 10K lines of code by the AI-generated WebGCS in about 100 hours over a course of 2.5 weeks. Burke ran several development 'sprints' using different AI models and coding tools, with each tool helping add new features step by step until the full system was working.
In sum, Burke’s AI-generated drone software represents a pioneering step in autonomous system design, combining generative AI’s coding power with embedded hardware for fully self-hosted and flexible drone control, reshaping how autonomous machines may be developed and function in the future.
- The AI-generated drone software, such as ChatGPT and others, is revolutionizing robotics by automating the creation of a drone's control system, a feat that was previously time-consuming using traditional human coding practices.
- This project, which incorporates robots at two levels—the AI models generating the code and the drone executing the AI-authored software autonomously—represents a significant leap towards robots co-designing and building other robots' cognitive systems.
- The use of generative AI for the development of autonomous systems also opens up opportunities in various fields, including spatial AI and aerial imaging, but it brings forth ethical, safety, and operational challenges due to the drones' ability to self-generate control logic and operate in unpredictable environments.
