IMPROV: Toolbox for Building Safe, Modular Robots Aware of Surroundings

Robots operating in manufacturing facilities have often posed risks to workers because they are not responsive enough to their surroundings. To make it easier for people and robots to work in close proximity in the future, researchers from the Technical University of Munich (TUM)  have developed the IMPROV system. The toolbox principle allows for the simple assembly of safe robots using various components. The modules can be combined in almost any way desired, enabling companies to customize their robots for a wide range of tasks – or just replace damaged parts.

Transcript

00:00:00 The concept of interconnectable modules for self-programming and-self verification, “IMPROV”, ensures a quick reconfiguration process of modular robots. It allows for stable and accurate control as well as verified safety for humans despite different robot geometry. The concept is shown on a commercially available set of modules that have standardized interfaces. Furthermore we created our own modules through 3d printing. Each module has information on its dynamics, kinematics and geometry. Users can then build the modular robot according to their needs. A centralized controller collects the data after the assembly process. The data is then used to automatically create the kinematic and the dynamic models which are used to parameterize model based controllers.

00:00:55 The robot geometry is also automatically created and this is used for collision checking and self-verification. We ensure human safety through an online verification approach. Here we show that by using trajectory scaling to stop the robot, a collision can never occur when the robot is moving. Only trajectories that allow for reaching a safe stop before a potential collision with humans are passed to the tracking controller. For this we require a prediction of the future occupancies of both the robot and the human. The robot's occupancy is obtained from its trajectory and the automatically generated geometry. For the human's occupancy we created a motion model which, based on its current pose, captures all possible future movements of the human body in a conservative way. The self-verification works for any module composition without manual reprogramming. As an example, it works for a six degree of freedom configuration as well as for an eighth degree of freedom configuration.

00:02:01 Comparisons with static safety zones revealed that the robot uses 36 percent less time to fulfill its task.