We say a lot with our hands. We spread them wide to indicate size, stab the air for emphasis and reach out to draw people in. Waving our hands about when we speak makes us appear less robotic 鈥� and that鈥檚 true for robots听迟辞辞.
Youngwoo Yoon at the Electronics and Telecommunications Research Institute in Daejeon, Korea, and his colleagues trained a machine learning system to match hand gestures to different words and phrases by showing it 52 hours of TED talks taken from 1295 videos on YouTube. They then tested the software in a humanoid robot.
Some humanoid robots, such as Pepper or RoboThespian, move their hands when they talk but these actions are typically hard-coded by humans, which takes time. The robot鈥檚 repertoire is also limited to the gestures its designers have given it and the actions soon feel repetitive.
Advertisement
To avoid this, the researchers let their software learn its own gestures by watching people speak. For each frame in the videos, they extracted the speaker鈥檚 pose 鈥� including the position of the head, neck and shoulders, as well as arms 鈥� and mapped this onto what the speaker was saying at the time.
The resulting system was able to generate gestures for words and phrases for any length of speech, including a crooked arm to suggest holding something, open arms to suggest inclusiveness and pointing gestures for 鈥測ou鈥� and 鈥渕e鈥�.
To test the system, the researchers recruited 46 crowdworkers, aged between 23 and 70, who were all native or proficient English speakers, from Amazon鈥檚 crowdsourcing website Mechanical Turk.
These participants judged the learned gestures to be more human-like, more friendly and a better match with the words spoken than those produced by existing approaches. Although there is still some way to go before the robot gestures are fully convincing. In the future, the team plan to personalise the hand movements so that robots don鈥檛 all use the same gestures.
Alan Winfield, a roboticist at the University of the West of England, says that body language is an incredibly important part of human interaction. He and his colleagues have developed a robot that mimics the facial expressions of the people it talks to. 鈥淎 very high percentage of communication is through gestures,鈥� he says.
Reference:
