The next IPAB workshop will take place 17/10/2019 at 12:45 in room G.03 (pastries will be served). Hakan Bilen will be speaking.
Title: Two ways to enhance the robustness of quadrupedal robots.
Summary: In recent years, the hardware and software of quadrupedal robots have progressed to a point where the basic-functionality of such robots can be performed reliably. However, when encountering challenging terrain or scenarios with disturbances (i.e. the expected use cases for these robots) their robustness requires further attention. In this talk, I will discuss my work on two ways to enhance the robustness of these robots. First, I will discuss planned contact between the robot body and ground: can we use such contact to help stabilize the robot against disturbances?
Second, I will discuss how to deal with unplanned contact between the ground and the robot has occurred: how can we train a robot to recover from falls?
Title: Is it possible to enhance robots dynamic dexterity?
The transition of robots from highly controlled to unstructured environments has highlighted the limitations of available controllers in handling uncertainties and nonlinearities. On the other hand, biological systems naturally achieve exceptional performances under the same conditions, and they are a valuable source of inspiration to improve the robots' controllers. A typical example is the application of the impedance controller, which has been used to study human reaching movements and lead to Passive Motion Paradigm (PMP) theory. A theory that works well for movements dominated by body dynamics, but it does not capture movements where the external dynamics cannot be neglected. The extension of the PMP to such cases requires the identification of a model that can explain how environmental dynamics affect motor planning. Available frameworks all require an accurate a priori knowledge of the system dynamics evolution, which is not obtainable in highly dynamics situation cha!
racterised by a limited horizon of events. However, recent studies have shown that human learning of external dynamics is quantifiable by their ability to synchronise with the environmental dynamics; thus, taking advantage of its energy flow. Starting from such results, we have been developing a modified attractor for the impedance controller that allows spontaneous synchronization with external dynamics. The proposed controller relies on passivity and a fractal topology to guarantee instantaneous adaptation to changes of environmental dynamics without the need of any accurate model of the environmental interactions, which is supported by the early results obtained with a 7-DOF arm. We are currently pursuing the extension of the proposed controller in applications such as teleoperation and locomotion, where it is not possible to obtain accurate a priori knowledge of the environmental interactions. Furthermore, we are also working on the integration of the controller with the PMP framework and the development of an experimental set-up to study possible correlation with human motor control.