Robohub.org
 

Quadrocopters learn from prior experience to improve slalom flying


by
11 June 2013



share this:
QuadrotorSlalomLearningPriorExperienceSmall

A new video released today by researchers from the Flying Machine Arena shows how a quadrocopter is able to learn from prior experience to improve future performance.

This new research is an extension of results published last year by the same group, which show how quadrocopters can learn to fly high-performance slalom courses (video).

Much like humans learn through repetition and practice, the quadrocopter repeatedly flies the slalom course, records any errors made and then tries to compensate for these errors during the next attempt.

However unlike human-skill-learning, the experience gained by flying one slalom course, was not carried over to a new course – these had to be learned from a state of zero experience.

A new video released by the group today (embedded below), extends this research and enables the quadrocopter to learn from prior experience.

By using knowledge of previously flown slalom courses, the quadrocopter is able to improve its performance on unseen courses, to the point where they can be accurately flown the first time, without practice.

Full Disclosure: I am the lead author of the video and associated research.



tags: , , , , ,


Mike Hamer





Related posts :



Robot Talk Episode 136 – Making driverless vehicles smarter, with Shimon Whiteson

  05 Dec 2025
In the latest episode of the Robot Talk podcast, Claire chatted to Shimon Whiteson from Waymo about machine learning for autonomous vehicles.

Why companies don’t share AV crash data – and how they could

  01 Dec 2025
Researchers have created a roadmap outlining the barriers and opportunities to encourage AV companies to share the data to make AVs safer.

Robot Talk Episode 135 – Robot anatomy and design, with Chapa Sirithunge

  28 Nov 2025
In the latest episode of the Robot Talk podcast, Claire chatted to Chapa Sirithunge from University of Cambridge about what robots can teach us about human anatomy, and vice versa.

Learning robust controllers that work across many partially observable environments

  27 Nov 2025
Exploring designing controllers that perform reliably even when the environment may not be precisely known.

Human-robot interaction design retreat

  25 Nov 2025
Find out more about an event exploring design for human-robot interaction.

Robot Talk Episode 134 – Robotics as a hobby, with Kevin McAleer

  21 Nov 2025
In the latest episode of the Robot Talk podcast, Claire chatted to Kevin McAleer from kevsrobots about how to get started building robots at home.



 

Robohub is supported by:




Would you like to learn how to tell impactful stories about your robot or AI system?


scicomm
training the next generation of science communicators in robotics & AI


 












©2025.05 - Association for the Understanding of Artificial Intelligence