Scientists Develop System to Help Robots Gesture Without Losing Balance

With a new control system developed by German researchers, robots can be made to move with expression and react to what is around them, all while keeping their balance. 

Till Hielscher and Kai O. Arras of the Socially Intelligent Robotics Lab at the University of Stuttgart are the ones behind the technology, called RoboMMT, as they describe in their research paper exploring how a robot might employ body language yet stay physically stable. 

The system was evaluated using two humanoid robots and a quadruped, with physical testing conducted on a Unitree Go2 quadruped. 

Teaching Robots to Move Naturally 

There is a growing tendency to build robots for work that involves some measure of human interaction, but even an unassuming gesture can be problematic for a machine obliged to always hold its balance. A humanoid that alters the pace of a movement or turns to face another will have its weight shifted, compromising stability. 

RoboMMT is a solution to that; it has a controller that makes constant adjustments so that the robot’s motion is both expressive and steady. The system gives researchers the ability to put their own spin on a gesture’s direction, timing or intensity within the confines of the robot’s physical make up. One can also have the robot direct a gesture to a particular spot or turn to a person in motion. 

Testing Balance Across Different Robots 

To put the system through its paces, the team used a pair of humanoids and a quadruped, doing hands on testing with the Unitree Go2. In simulations of the Unitree G1 and OP3 humanoids and the Go2, the controller completed 99.3% or more of the 2,048 motion sequences put before each robot. The interactive gesture system’s spatial targets were not part of those tests though. 

The technology was moved over to an actual Unitree Go2, with no extra training required. One experiment had the robot complete more than 90 seconds of uninterrupted movements under external duress; the sequence was repeated five times without the robot falling whatsoever. 

While the physical side of things has been limited in demonstration, the results point to a capacity for robots to be stable and expressive at the same time.  

Making Human-Robot Interaction More Natural 

Down the line, this may well be of use to robots operating in public or healthcare settings where one needs to convey intention and attention without having to put it into words. 

But for now, RoboMMT is experimental. No conclusions have been drawn by the researchers as to whether people would find the robot’s mannerisms any more engaging or easy to read. 

And to see how reliable the technology is in less controlled surroundings, more testing is required. It is an issue of social robotics in the end: helping machines communicate through movement without compromising their ability to remain safely balanced.