AI Drone faster than Humans? Time-Optimal Planning for Quadrotor Waypoint Flight

AI Drone faster than Humans? Time-Optimal Planning for Quadrotor Waypoint Flight

Quadrotors are among the most agile flying robots. However, planning time-optimal trajectories at the actuation limit through multiple waypoints remains an open problem. This is crucial for applications such as inspection, delivery, search and rescue, and drone racing. Early works used polynomial trajectory formulations, which do not exploit the full actuator potential because of their inherent smoothness. Recent works resorted to numerical optimization but require waypoints to be allocated as costs or constraints at specific discrete times. However, this time allocation is a priori unknown and renders previous works incapable of producing truly time-optimal trajectories.
To generate truly time-optimal trajectories, we propose a solution to the time allocation problem while exploiting the full quadrotor’s actuator potential. We achieve this by introducing a formulation of progress along the trajectory, which enables the simultaneous optimization of the time allocation and the trajectory itself. We compare our method against related approaches and validate it in real-world flights in one of the world’s largest motion-capture systems, where we outperform human expert drone pilots in a drone-racing task.

Reference:
P. Foehn, A. Romero, D. Scaramuzza
“Time-Optimal Planning for Quadrotor Waypoint Flight”
Science Robotics, July 21, 2021
PDF: http://rpg.ifi.uzh.ch/docs/ScienceRobotics21_Foehn.pdf

For more info about our research page on :
1. Drone Racing: http://rpg.ifi.uzh.ch/research_drone_racing.html
2. Agile Drone Flight: http://rpg.ifi.uzh.ch/aggressive_flight.html

Affiliations:
All the authors are with the Dept. of Informatics, University of Zurich, and Dept. of Neuroinformatics, University of Zurich and ETH Zurich, Switzerland http://rpg.ifi.uzh.ch/

Music Credits: scottholmesmusic.com under Free Creative Commons License

35 Comments

  1. edward Burgos on December 30, 2021 at 12:23 pm

    Apagué y Vamos 😭😭😭😭

  2. MrHangrun on December 30, 2021 at 12:26 pm

    Background Music much to loud.

  3. Patfish on December 30, 2021 at 12:27 pm

    Das ein Computer gesteuerte Drohne innerhalb einer Mocap-Area eine Ideallinie abfliegen kann ist doch keine große Kunst?… Autonomie bedeutet für mich, wenn eine Drohne wirklich selbständig außerhalb einer Motion Capture Area auf dynamische Hindernisse reagiert und jene dann auch optimal umfliegt.

  4. Eric Stratten on December 30, 2021 at 12:31 pm

    Where the human pilots handicapped by carrying added weight required by the computer controlled version? If you let the humans choose their own optimized quad-copter is the computer controlled version still capable of beating them?

  5. giantasparagus on December 30, 2021 at 12:33 pm

    damn

  6. SAMET ELMACI on December 30, 2021 at 12:33 pm

    omg

  7. LCH - P4X on December 30, 2021 at 12:34 pm

    Algo is NOT an AI

  8. Beep Tube on December 30, 2021 at 12:35 pm

    Which racing pilots were flying the track?

  9. Ayla Rivers on December 30, 2021 at 12:37 pm

    They need to delete that algorithm. Seem like we’re always getting closer to the skynet outcome and the one that merge with ai

  10. djmips on December 30, 2021 at 12:37 pm

    To the researchers: The video orange highlight was merely obscuring detail not helping.

  11. SJChannel on December 30, 2021 at 12:40 pm

    There is one aspect of the comparison between your autonomous system and the human pilots that seems a bit unfair. A human FPV pilot must be able to see where he’s going; i.e., he must keep the quadcopter oriented such that the camera faces in the direction of travel. Your autonomous system does not have to waste time on that. It can fly equally well forward, backward, or sideways. It would be interesting to see how the performance of your system would be affected by an additional constraint that the quadcopter must always be oriented with its (non-existent) camera facing forward.

  12. Bez on December 30, 2021 at 12:40 pm

    Does the system handle dynamic events such as a gate moving in the wind? Or even blowing over completely.

  13. LeeWhitcher on December 30, 2021 at 12:42 pm

    Nice work Phillipe and team! Computer vision really is the limiting factor in optimality, now… Hurry up, CS folks!

  14. macman90 on December 30, 2021 at 12:43 pm

    Optimal pizza delivery times!

    No joke, it looks great.

  15. MCK FPV on December 30, 2021 at 12:43 pm

    Wanna race

  16. Stefan Steiner on December 30, 2021 at 12:44 pm

    Nice!

  17. drone racer on December 30, 2021 at 12:45 pm

    In 1997 the first computer beat a chess grand master, humans will never regain the title, and the same will apply to drones from today.

  18. Bez on December 30, 2021 at 12:49 pm

    How much time is allocated to the computation of the trajectory preflight?

  19. Duc Nguyen on December 30, 2021 at 12:51 pm

    Great!

  20. Bez on December 30, 2021 at 12:51 pm

    Can the system handle multiple pilots on the track simultaneously?

  21. Frank Sarfino on December 30, 2021 at 12:51 pm

    Who were the pilots? Inquiring minds want to know

  22. Gabe Herbertson on December 30, 2021 at 12:52 pm

    LAME! When the price point is equal, only then can you call it a race!

  23. NtHwk_ on December 30, 2021 at 12:52 pm

    im guessing at this rate, the hardware is gonna be the limiting factor

  24. Paleo Geology on December 30, 2021 at 12:53 pm

    Yeah BUT can they do this when a course is changing such as outdoors over various landscapes? Probably not

  25. The Max on December 30, 2021 at 12:56 pm

    *Alex Vanover entered the chat*

  26. Hari Krishnan on December 30, 2021 at 1:01 pm

    very cool

  27. Vadim Romanovich on December 30, 2021 at 1:02 pm

    Looks like time trials only. Would be interesting to see how it handles head to head competition or against a number of pilots.

  28. bernard Li on December 30, 2021 at 1:02 pm

    https://www.youtube.com/watch?v=MvRTALJp8DM UPENN has done similar research in movement. The problem with this is not the possible algorithm (although it’s definitely impressive!) but the challenge is having state estimation fast enough to keep up with the speeds of pilots. Until that challenge is solved, basically any university can create their own algorithm to try (and we’re working on it in Texas :D).

    It’s good to see UZH keeping pace after their Alphapilot attempt! Let us know if ya’ll are attending any robotics conferences!

  29. Pratik Prajapati on December 30, 2021 at 1:03 pm

    Its mind blowing

  30. Strages Powers on December 30, 2021 at 1:03 pm

    welcome to skynet

  31. ShiftFPV-Proximo on December 30, 2021 at 1:05 pm

    This is really awesome but I don’t know why drone racing would end up being ai because the competition is between humans

  32. yalmadiable on December 30, 2021 at 1:07 pm

    Well first you are giving the drones more power than human where they use the IR cameras indoor to know there position in space accurately from all angles at all times. The question is how feasible would such a task will be if the drone is outdoor and flown by Jetson using a wireless data link?

  33. Spencer Drager on December 30, 2021 at 1:07 pm

    Incredible. I’m curious how it would handle higher TWRs. I imagine it could start coming up with some wacky routines that would really blow a human operator out of the water due to how difficult they come to control (due to human response time and input resolution).

  34. NS Rana on December 30, 2021 at 1:09 pm

    wow, thats amazing, I am interested to know more about it.

  35. overpropped rob on December 30, 2021 at 1:22 pm

    Thats impressive. Thanks for the insights 👍

Leave a Comment