Drone-racing champions outpaced by AI

Because the state variables encode a higher level of abstraction than the raw images, the reinforcement-learning simulator does not need to render a rich visual environment. Given good state estimation, this reduces the differences between the simulated and real systems considerably. [...]it leads t...

Full description

Saved in:
Bibliographic Details
Published inNature (London) Vol. 620; no. 7976; pp. 952 - 954
Main Author de Croon, Guido C. H. E.
Format Journal Article
LanguageEnglish
Published London Nature Publishing Group 31.08.2023
Subjects
Algorithms
Artificial intelligence
Cameras
Engineering
Neural networks
Robots
Simulation
State estimation
Vision systems
Visual stimuli
Online AccessGet full text

Cover

More Information
Summary:Because the state variables encode a higher level of abstraction than the raw images, the reinforcement-learning simulator does not need to render a rich visual environment. Given good state estimation, this reduces the differences between the simulated and real systems considerably. [...]it leads to much faster simulations, enabling the system to learn in around 50 minutes. [...]they will help all robots, whether they are used for driving, cleaning or inspecting, to get more out of their limited onboard resources.
Bibliography:SourceType-Other Sources-1
ObjectType-News-1
content type line 66
ISSN:0028-0836
1476-4687
DOI:10.1038/d41586-023-02506-8