Researchers including Sebastien Lengagne, Karim Bouyarmane, and Abderrahmane Kheddar at AIST are trying out a novel approach to navigating cluttered environments. Rather than seeing household objects as obstacles to be avoided, they are programming the robot to use them for support. For example, if a human were to look under a desk they’d probably hold onto it for support, but a robot is usually programmed to do these sorts of tasks without touching anything. In physical tests they’re keeping things relatively simple; the HRP-2 Promet can steady itself on a table when sitting down in a chair or when taking a long step. In simulation they’re able to get a bit more ambitious, like having two robots cooperatively manipulate an object, or climb a steep incline.
Collisions with objects, including other robots, and attempting to balance on them together pose a number of problems. For example, a robot needs to know what objects can physically support them in a given task. A robot’s joints could be damaged or the robot may slip if they aren’t properly supported. So the researchers are developing simulation software that determines contact points between the robot and any other object (even other robots) and the environment, while also factoring in the effects of gravity and actuator torques, ensuring they are stable. The resulting work could also be used to help robotic manipulators grasp objects, or even work together to manipulate objects too large for one robot to carry alone.