2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
American engineers have developed an algorithm for bipedal robots that allows them to use nearby surfaces to maintain balance. For example, if a robot near a wall is pushed, it will quickly calculate the optimal angle between body parts and rest its hand against the wall, the developers say in an article presented at the ICRA 2018 conference.
Bipedal robots are attractive to engineers because their potential, for example, cross-country ability, is higher than that of wheeled or tracked robots. Nevertheless, while they are at an early stage of their development, even many advanced developments, including those participating in the DARPA Robotics Challenge, do not cope well with non-standard situations, such as side shocks or a stone slipping out from under their feet. The fact is that most bipedal robots use quasi-static movements, in which the projection of the center of mass of the robot is always inside the leg-to-surface contact patch. Because of this, they go out of balance with noticeable deviations and fall.
As a solution to this problem, engineers offer two different approaches, similar to how people maintain balance. One of them involves the use of dynamic movements - then the robot maintains balance, quickly rearranging its legs. Another approach implies that the robot can use not only its own body parts, but also surrounding objects, such as walls or handrails, to save itself from falling. Shihao Wang and Kris Hauser from Duke University took the second approach and taught the bipedal robot to quickly analyze the environment and use objects from it as a support.
The algorithm created by the engineers solves three problems: detects a fall and predicts its direction, uses the robot's arm to stop the fall and, if the arm is not extended too far, allows the robot to push off and return to its original position. The developers have simplified the model describing the robot to three rigid sections connected by two joints. This made it possible to adapt the algorithm to work in real time. In addition, the algorithm works correctly only when falling straight sideways or forward without twisting and other movements.
The robot model used in the work
The engineers used a ready-made robot ROBOTIS Darwin Mini in their work, complemented by a Raspberry Pi 3 single-board computer, a nine-axis inertial unit and a touch sensor mounted on the end of the arm. In the current version of the robot, engineers did not install a camera or other sensor and pre-loaded data about the surrounding surfaces and the distance to them. After the algorithm detects a fall, it analyzes the position of the robot relative to the surfaces in the direction of the fall and selects such a relative position of the legs, body and arms that will allow it to lean on the surface and not fall after that. After stopping, the robot uses the same model and analyzes if it can bounce off the surface and return to its original position.
The engineers said that in the future they plan to install cameras in the robot and analyze the environment in real time without using preliminary data. In addition, they plan to improve the algorithm for detecting falls while walking.
In 2015, another group of American engineers created an algorithm that allows bipedal robots to soften the impact of a fall. It determines the proper sequence of contact between the parts of the structure of the robot and the surface in order to minimize the force of the impact. Moreover, in some cases, the robot not only stretches its arms forward, but also rolls over its back.