Why does the simple act of making a bed become a significant robotics achievement? The primary feature of Figure’s latest demo video involves two humanoid robots entering the bedroom and cleaning it within two minutes, putting things away and making the bed. The key behind it is much less visible at first glance since making the bed seems like a fairly simple process that most people do routinely and effortlessly.
This exact fact makes such a sequence important making the bed becomes easy precisely because humans intuitively deal with all kinds of forces involved in the process. A comforter is not an inflexible element but something that moves, folds, and changes position depending on what happens to the edges. For many years, cloth manipulations became one of the crucial elements demonstrating how well robots could cope with dynamic environments. This point is especially important in light of recent developments in laundry automation reported in Smithsonian Magazine.
As was mentioned by this publication, there are thousands of ways to fold a cloth piece, so every attempt to interact with the object requires a new strategy for grasping and positioning the material. Figure’s system deserves attention for this reason precisely since this particular company claims that no teleoperations are involved in controlling the robot.
According to their presentation, the system features one neural network running everything from pixels controlling robot locomotion and manipulation at the same time. This approach has an essential advantage when it comes to manipulating things in the room the machine will need to change location, move the arms simultaneously, pick things up and release them at the same time. The bedroom demonstration builds upon the same principles as previous work on living-room cleanliness featuring one humanoid robot.
In the living room, the robot moves around, sprays a towel, picks up containers from the floor, holds them while reaching some place with one arm, tucks the objects under the arm and finally uses its ability to rotate the remote control until hitting the right button. In this context, Figure states that living-room cleaning poses a challenge for its robots due to unpredictable placement of objects and the need for manipulations in narrow spaces. Bedroom cleaning becomes a next step involving multiple robots working in the same environment.
It is even claimed that the system allows the machines to interact with each other without a central controller, shared planner or direct communication. In the practical context of this problem, the robots monitor the situation and adjust the process depending on what happens during manipulations. According to Corey Lynch, who works for Figure as a director of AI, “To be clear, there’s no explicit messaging between these robots, they coordinate their actions fully visually, e.g. head nods”. This fact makes the process much more complex than picking up the objects.
The current robotics market features a number of alternative projects. UniX AI’s humanoid Panther has a wheeled body and demonstrates the choice faced by manufacturers between bipedal and stable designs. However, Figure aims at something else developing a universal system capable of performing a range of tasks in real life without the help of programmers.