Dancing droids: Scientists teach humanoid robot sweet moves, high-fives, hugs

Aiming to reshape public perception of robots, a group of engineers have trained a humanoid to learn and execute a variety of expressive actions.

The University of California researchers taught the robot how to perform basic dancing moves and hand motions, including waving, high-fiving, and hugging, while keeping a steady stride over various surfaces.

According to the team, the enhanced expressiveness and agility of this humanoid robot open new avenues for improving human-robot interactions in various settings.

These include factory assembly lines, hospitals, and homes, where robots could safely work alongside humans or replace them in hazardous environments like laboratories or disaster sites.

“Through expressive and more human-like body motions, we aim to build trust and showcase the potential for robots to co-exist in harmony with humans,” said Xiaolong Wang, a professor at UC San Diego Jacobs School of Engineering, in a statement .

Unique training module

The objective of the study was to enable humanoid robots to generate rich, diverse, and expressive motions in the real world. The proposal involves learning a whole-body control policy for a human-sized robot to mimic human motions as realistically as possible.

To train the robot, researchers use large-scale human motion capture data in a reinforcement learning framework. However, directly applying this data to the robot doesn’t work due to differences in movement abilities.

The team’s Expressive Whole-Body Control (ExBody) solves this by making the robot’s upper body imitate human motions while allowing its legs to simply follow a given speed.

According to researchers, the humanoid robot’s expressiveness comes from its training in a wide range of human body movements, allowing it to effortlessly mimic and generalize new motions. Similar to a quick-learning dance student, the robot swiftly learns new routines and gestures.

The team trained the robot using an extensive collection of motion capture data and dance videos. Their unique technique trained the robot’s upper and lower body separately.

According to the team, the robot’s upper body was trained to replicate various reference motions, such as dancing and high-fiving, while its legs focused on maintaining balance and navigating different terrains with a steady stepping motion.

“The main goal here is to show the ability of the robot to do different things while it’s walking from place to place without falling,” said Wang.

Enhanced robotic capabilities

The robot’s whole structure is governed by a single policy, even though its upper and bottom bodies were trained separately. The coordinated policy guarantees the robot’s ability to walk steadily on surfaces such as grass, gravel, dirt, wood chips, and sloping concrete roads while also enabling it to execute intricate upper-body movements.

A virtual humanoid robot was used for simulations before being moved to a real robot. The robot showed that it could perform both novel and trained actions in practical settings.

At the moment, a gaming controller used by a human operator controls the robot ‘s speed, direction, and certain maneuvers. In the future, the team plans to add a camera to a later model so that the robot can operate independently and navigate different types of terrain.

They are also now concentrating on refining the robot’s design to handle more intricate and detailed tasks. “By extending the capabilities of the upper body, we can expand the range of motions and gestures the robot can perform,” said Wang.

According to the study abstract, the team believes their “method paves the way for the development of reliable and versatile humanoid robots, capable of performing multiple functions effectively.”