Learning from the past, drone navigates course with ease

“Humans learn from experience: when we try something and fail, we try doing it a different way the next time around. And we are incredibly efficient at this process.” Making mistakes can be an essential step in becoming aware of our progress. They allow us to become aware of our actions and in turn, learn from them, and adjust accordingly. Self-awareness of personal experience is a key to improvement; this insight, combined with an understanding of how we learn, can be greatly beneficial to learning optimization. This philosophy is at the heart of the innovative research being performed by Angela Schoellig and colleagues from the Swiss Federal Institute of Technology Zurich. The goal of this research is to develop algorithms that optimize the learning ability of autonomous systems based on human learning models. These systems would essentially provide robots with learning algorithms that act as a self-awareness feedback loop running off of its past experience data.

Schoellig and company’s optimization-based iterative learning models have been recently tested on the flight trajectories of quadrocopters. Slalom poles are placed arbitrarily placed in an open space to create an obstacle course for the drone to navigate. First, an optimized flight trajectory is calculated around the series of poles. The quadrocopter is then fed this data and attempts to follow the pre-computed flight path; due to unpredictable aerodynamics, the drone will initially fail to execute and hit several poles. However, thanks to the use of optimal filtering methods combined with convex optimization techniques, a model is created to help the drone iteratively learn the prescribed path through practice! The deviation from each experimental run is compared to the desired flight path, is corrected for, and is fed back into the input of the quadrocopter. This open-loop learning model allows for the vehicle to detect recurring errors and quickly correct for them after each run.

It is quite possible this research can have a great impact on future development of self-learning robotics. Here’s an interesting idea: harness the power of the imagination and, in addition to the experiential learning ability, supply these drones with visualization skills. In other words, allow the mini-copters to run simulations prior to each practice run before its normal iterative learning process. Should make for an interesting addition to the already human-based learning models used in Schoellig’s research.

I love first-person views of flying robots. This one especially. See the last 30 seconds.

You don't have permission to edit metadata of this video.

Edit media

Dimensions x Small Medium Large Custom

Subject (required) Brief Description Tags (separated by comma) Video visibility in search results Visible Hidden

Parent content

Learning from the past, drone navigates course with ease

Poster

Upload Preview

Ok

Cancel

Cabe

http://twitter.com/Cabe_e14