M. Frietsch

Semi-autonomous flying robot for physical interaction with environment

This contribution presents the first results of the development of an unmanned aerial vehicle (UAV) which is capable of applying force to a wall while maintaining flight stability. This is a novel idea since UAVs are used so far only for tasks without physical contact to the surrounding objects. The basis for the work presented is a quadrotor system which is stabilized with an inertial measurement unit. As a new approach an additional actuator was added to generate forces in physical contact while the UAV stays horizontal. A control architecture based on ultrasonic distance sensors and a CMOS-camera is proposed. The performance of the system was proved by several flight tests. Potential applications of the system can be physical tasks at high places like cleaning windows or walls as well as rescue or maintenance tasks.

A new two-layer reinforcement learning approach the control of a 2DOF manipulator

This paper presents a new machine learning approach, based on reinforcement learning, to control a highly nonlinear robot demonstrator. Learning is achieved using an on-policy temporal difference learning agent framework. The agent controls the movements of the robot by choosing torques for each joint and immediately receives a feedback signal. The implemented SARSA-agent uses an update rule that calculates the estimate of the current state-action by using the current state-action value, the received reward and the following state-action value. To handle the high number of state-action value pairs, a hash-table is used to efficiently store and access these values inside the lookup-table. To accelerate the learning process of more complex motions, a new second layer approach is introduced. In this approach, a library of simple motions is created in the first layer. The second layer-agent then combines the gathered experiences to achieve a faster solution for more complex motions. The evaluation of the two-layer agent shows that the combination of both layers dramatically increases the speed of finding a solution. Additionally, its solution is often better than the solution found by the pure reinforcement learning agent.

A new bionically inspired approach to increase positioning accuracy of robotic systems

In state of the art robotics, high positioning accuracy is achieved by using solid and stiff components as well as powerful drive units which have no backlash. In contrast, human beings are able to achieve remarkable high positioning accuracy despite of low mass, low power consumption and relatively simple mechanics. One approach to obtain this accuracy is to temporarily create additional supporting structures by interacting with the direct environment, e.g. supporting the heel of the hand on a table for writing. This article deals with the essential idea of applying this method correspondingly into the field of robotics. It points out advantages and disadvantages as well as possibilities to realize this method in different scenarios. With simplified conditions, the influence of propping up on the stiffness and hence on the positioning accuracy is examined using different simulation models. It turns out that blocking of even one degree of freedom in one direction, can lead to a significant improvement regarding stiffness and therefore positioning accuracy. This approach could be used in diverse applications e.g. deburring with an industrial robot or in a humanoid robot to increase the reliability of a process or to reduce cost of components.

Improving positioning accuracy of robotic systems by using environmental support constraints: a new bionic approach

In state of the art robotics, high positioning accuracy is achieved by using solid and stiff components as well as powerful drives. But in the field of social robotics, for example humanoid robots, it is often not possible using this approach due to special boundary conditions like design-space, weight-limitations, power-storage and many more. By contrast human beings are able to achieve remarkable high positioning accuracy despite of low mass, low power consumption and relatively simple mechanics. One approach to obtain this accuracy is to temporarily create additional supporting structures by interacting with the direct environment, for example supporting the heel of the hand on a table for writing. This article deals with the essential idea of applying this method correspondingly into the field of robotics. Using different simulations the influence on stiffness and positioning accuracy is examined. It turned out that blocking of even one degree of freedom can lead to a significant improvement regarding stiffness and therefore positioning accuracy.

A Two Layered Process for Early Design Activities Using Evolutionary Strategies

Especially in system and conceptual design activities designers resort to already existing components that are combined and arranged to a new system which has to fulfill a predefined set of requirements. Designers have to deal with requirements and constraints that are changing during the development process. General goal is an automatic generation of compatible conceptual design proposals that meet the predefined requirements such as design space, EMC, etc. To this, libraries containing standardized data on components have to be developed. These libraries include component specific characteristics and data such as CAx models or efficiency factors. In an iterative process compatible systems are configured and evaluated by means of CAx based analyses. Afterwards an optimization system based on genetic algorithms accesses these data to find optimal configurations. By combining the optimization algorithm with a CAD system, design proposals are directly visualized and can be processed by the designer in the further product development process.