American and Italian engineers have taught a one-legged robot Salto-1P after the jump to land at the planned point with great accuracy and to stop without falling. Previously, the robot could only jump without stopping. The developers have submitted an article about the new abilities of the robot at the conference ICRA 2020 (Preprint available on the University website) and told about them in an interview with IEEE Spectrum.
Salto is a one-legged hopping robot, the first version of which the American engineers introduced in 2016 and then modified. A modern version of the Salto-1P controls your jumps and landing area with three-section legs with electric motor in the base, a flywheel on one side and two side screws. This allowed the robot to accurately count your jumps and jumping over objects with different height. But early to maintain stability it was necessary to constantly jump and stop him or were caught with your hands in the air, or sent to the grid.
In the new version of the control algorithm engineers under the leadership of Ronald Feringa (Ronald Fearing) at the University of California at Berkeley greatly increased the precision landing of a robot and taught him to stop after that and not to push and jump on. The main improvements are in control of angular momentum. For this, the robot accurately calculates and adjusts wheel angle of its inclination to the moment of landing, so that after touching the floor the moment the pulse was small and sufficient to ensure that the robot is moved to a vertical position. If the angle at the moment of touching the floor will be less than desired, that is, the robot will be closer to the vertical position, angular momentum is too large, so that the flywheel could to extinguish it, and the robot spun and fell further along the trajectory.
In addition, the developers have improved the calculations of the level of modpluginevent feet
at the moment of landing, as well as the calculation of the angle at the beginning of the jump. Last
the item is important, because after separation the robot can only control
its rotation, and the trajectory of the center of mass it is already unable to change,
therefore, the landing point depends on the initial phase.
Improvements in the calculation of the initial phase of the jump helped to reduce the standard deviation from the estimated landing point from 9.2 to 1.6 cm. And the best control over the tilt angle when landing allowed the robot to stop and execute a few precise jumps in a row, which would be difficult to achieve without a possibility of stopping.
In this work, the robot to carry out calculations, drawing not only on his testimony, but also on the exact appearance a system of motion capture. But last year the developers have created a completely independent version of the robot and showed video on which it is several hundred jumps in a row moves down the street.