A team of AI and robotics researchers from Carnegie Mellon University, USA, together with colleagues from NVIDIA, have developed a new model to teach robots natural movements. Previously, this was extremely difficult to do, but the scientists found a new approach. The results of their work were shown in a video.
A team of AI and robotics researchers from Carnegie Mellon University, USA, together with colleagues from NVIDIA, have developed a new model to teach robots natural movements. Previously, this was extremely difficult to do, but the scientists found a new approach. The results of their work were shown in a video.
The robots have learned to move well, but none of them, the team notes, do so with particular grace; they lack the fluidity and athleticism that are hallmarks of natural animal movement. The answer, they believe, was to focus on full-body training, Tech Xplore reports .
In their quest to develop a full-body workout, the team found that existing training models lacked adaptability and often used too many parameters, leading to over-cautious movements. This prompted them to develop a new two-stage model.
The first stage involves training an artificial intelligence module to understand videos of full-body human movements — with retargeting to key moments to take into account the capabilities of robots combined with motion tracking.
The second phase involves collecting real-world data to identify and reconcile differences between real-world actions (how people move in videos) and how robots might move. The result is a framework the team calls Aligning Simulation and Real Physics (ASAP).
To test the new framework, the researchers trained the robot to perform moves familiar to sports fans. The robot performed Kobe Bryant’s famous jump shot, LeBron James’ “Muffler” move, and Ronaldo’s mid-air spin. Each full-body skill was recorded as it was performed, and the results were posted on YouTube.
Looking at them, it's easy to recognize familiar movements and note the progress made in perfecting full-body movement. But it's also easy to see that there's still a lot of work to be done.
