Boston Dynamics robot seller said today that she expanded the capabilities of artificial intelligence (AI) for her human robot in Atlas, where he collaborated with the Toyota Institute of Research (TI) to enable the bilateral robot of the legs to learn new skills through human demonstrations, instead of updates of complex programs coding.
Specifically, the partners enabled the ATLAS model through a major behavior model (LBM) to give him new potentials that could have been a manually programmer, but it can now be added quickly and without writing a new code, by copying a human trainer. The partners said that the result allows the Atlas unit to perform a long and continuous sequence of complex tasks that require combining the processing of organisms with the movement.
Boston Dynamics makes commercial products such as dogs -like “Spot”, and is similar to dogs, and its “stretching” that combines a robotic arm with a mobile carrier to unload truck trailers independently. But it may be known for the Atlas design, which appeared in a series of famous videos that perform gymnastics and running through obstacles. Massachusetts does not sell this model commercially, but it is used by a Robot test bed.
In recent years, other sellers have added human robots to this product category, including the Apptronik’s Apollo numbers unit and Agfility Robotics. All of these robot models are currently used in the experimental tests of the business sectors, including logistical services and warehouses.
According to Boston Dynamics, its latest version of the ATLAS product allows the entire human body movements, such as walking, porn, and lifting, to accomplish a series of mobilization, sorting and organization tasks. He achieved these movements even when researchers interfered in unexpected physical challenges in the middle of the mission, such as closing the box cover and sliding it across the ground, which requires a atlas to respond to itself.
The two companies said that the result reaffirms the “amazing capabilities” of artificial intelligence techniques in developing human assistants for general purposes. “One of the main value proposals for humans is that they can achieve a large group of tasks directly in the current environments, but previous methods of programming these tasks cannot expand in the face of this challenge,” said Ross Tedrik, Senior Vice President of Big Behavior Models at the Toyota Institute for Research. “Large behavior models address this opportunity mainly – skills are quickly added through demonstrations of humans, and since LBMS increases strength, they require less and less demonstrations to achieve more more powerful behaviors.”
