Mark Laber, founder and chairman of Boston Dynamics, has given the world a zoo of two-legged and four-legged machines that can surprise parkour, infectious dance routines and hardworking stacks of shelves.
Raibert is currently trying to lead the revolution in robot intelligence and acrobatics. And he says recent advances in machine learning have accelerated the ability of robots to learn how to perform difficult movements without human help. “The hope is that we can create a lot of action without having to make everything that a robot does,” Raibert told me recently.
Boston’s dynamics may have pioneered the robotics of the foot, but it is now part of a crowded pack of companies offering robotic dogs and humanoids. Only this week, the startup, known as figures, showed off a new humanoid called Helix. Another company X1 showed off a muscular humanoid called Neogamma who does chores around the house. The third Apptronik said he plans to expand the production of his humanoid, called Apollo. However, the demonstrations can be misleading. Also, few companies have revealed the cost of humanoids, and it is unclear how many companies really expect to sell as home helpers.
The actual tests of these robots are quantities that can be independent of human programming and direct control. And it depends on the progress that Raibert advertises. Last November I wrote about my efforts to create an entirely new kind of model for controlling robots. If the work begins to pay off, we may see humanoids and quadrupeds progress more quickly.
Boston Dynamics sells four-legged robots called spots used in oil rigs, construction sites, and other places where wheels wrestle against the terrain. The company also manufactures humanoids called Atlas for research. According to Raibert, Boston Dynamics was able to upgrade Spot’s execution capabilities using an artificial intelligence technology called Renforce Learning, allowing it to move three times faster. The same method also helps Atlas walk with more confidence, says Raibert.
