Researchers at Georgia Tech developed a faster, cheaper way to teach a humanoid robot to walk on uneven ground. Machine learning PhD student Feiyang Wu led the work on a new whole-body controller. The team tested the robot on sand, soggy grass, gravel, slopes, stairs and level ground found on campus or simulated.
The method trains faster and uses less computing time. It worked well even on some surfaces that were not in training. Wu presented the training framework at the IEEE International Conference on Robotics and Automation.
The project was a collaboration between two schools at Georgia Tech and it can apply to other robots and tasks. It was supported by US research agencies.
Difficult words
- humanoid — a robot that looks or acts like a person
- controller — software or device that directs robot movement
- uneven — not flat, with different heights or slopes
- simulated — made to look or feel like the real thing
- collaboration — work done together by two or more groups
- framework — a set of ideas or plans for a project
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Which surface from the article do you think is hardest for a robot to walk on? Why?
- Why is faster training useful when building robots?
