The aim of this research activity is to develop model-based control strategies for tendon actuated continuum soft robots. Up until now, we have developed two tendon platforms which consist of joint-like modules with high dexterity and intrinsic mechanical robustness to absorb external impacts without harming the actuators. A planar mechanism which serves as a fundamental test platform to help the control methods to mature. And a spatial mechanism, currently being used as a neck of the humanoid robot David [1].
Existing modern control techniques, like a partial feedback linearization [2], have been successfully applied to both systems in the past for controlling their posture. In this respect, the tendon actuators have been assumed to be sufficiently fast and their dynamics are neglected in the approach. Structurally, an impedance-like controller is implemented which receives positions and velocities and generates forces. Since such model-based controllers require information of the systems state which are not measurable, a posture observer runs in parallel to the controller to provide this necessary information which are gained based on the actuation sensing [3]. In the above-mentioned directions, extensions or other modern control and observation techniques are mandatory to be investigated.
It is of primary interests that the developed methods for continuum soft robots are theoretically sound and work in practice.
Example of specific theses that are currently available can be found below. (at the time you are reading, they may very much be not up to date, but they can still serve as example)
Supervisors: Dr.-Ing. Bastian Deutschmann (contact data below), Dr. Cosimo Della Santina (cosimodellasantina[ @ ]gmail.com)
Additional possible supervisors: Dr.-Ing. Christian Ott, and Prof. Dr.-Ing. Alin Albu-Schaeffer.
References
So far, existing control approaches were implemented in two different ways: 1) The ultimate goal is to drive the robot to a desired configuration and the transient behavior is not considered. For this case, usually, inverse kinematics or inverse statics is applied to map the desired configuration into actuator space whereas the controller is closed within the actuators only [4]. 2) The ultimate goal is to move the robot along a desired trajectory. Here dynamic controllers are formulated in end-effector or configuration space assuming instantaneous action/reaction of the actuation system (i.e. the dynamics of the actuators is negligible) [5].
As a matter of fact, the actuator dynamics / the bandwidth of the low-level actuator controller cannot be neglected and usually results in a decreased performance and requires adaptation of the initially implemented control law of the simulations. This is specially in case true for approach 2).
This project aims to develop a trajectory-tracking control approach for a tendon-driven continuum type soft robot which is based on a model including the actuator dynamics. Promising directions are the well-known "backstepping method" [6] or a concept called "ESpi-control" [7].
The following steps are recommended:
Suggested as (but not limited to!):
References:
[4] Bajo, Andrea, et al. "Constrained motion control of multisegment continuum robots for transurethral bladder resection and surveillance." 2013 IEEE International Conference on Robotics and Automation
[5] Della Santina, Cosimo, et al. "Model-based dynamic feedback control of a planar soft robot: Trajectory tracking and interaction with the environment." The International Journal of Robotics Research 39.4 (2020): 490-513.
[6] Chalon, Maxime, and Brigitte d'Andréa-Novel. "Backstepping experimentally applied to an antagonistically driven finger with flexible tendons." (2014): 217-223.
[7] Keppler, Manuel, et al. "Elastic structure preserving (esp.) control for compliantly actuated robots." IEEE Transactions on Robotics 34.2 (2018): 317-335.
Soft robots of continuum type are expected to be applied as futuristic manipulators, humanoid joints or their equivalent prosthesis, or catheters for medical treatments. The developed systems in the community are diverse in terms of the shape and material used for the robot’s body, the applied actuation scheme and the robot's sensing capabilities. In the past years, the design and the mathematical description of such were discussed largely among the research community. To develop and implement model-based control in contrast, has not been addressed extensively in the past, however is in focus today.
Given a desired objective for the closed-loop system, a suitable controller is to be designed for continuum soft robots considering the relevant system-specific properties. A challenging topic which needs to be solved with care since the real performance will be visible when the developed algorithm is transferred and tested on the real hardware. Due to the diversity of the existing systems, these controllers are diverse as well and their reported performance is most often not directly transferable to other systems which limits their applicability. Standardized tests or metrics which assess the performance of the developed controllers objectively are missing.
In this work, this fundamental problem shall be addressed. The ultimate goal of this project is to develop and test a procedure based on meaningful metrics by which existing control approaches, e.g. [2] or [5], can be compared. Therefore, performance metrics are to be developed (for new metrics) / or implemented (if existing). To show the applicability of the developed metrics, three different model-based controllers are implemented on a common system (Fig. 1).
Thanks to their body elasticity, soft robots should be able to perform periodic tasks in an exceptionally good way.
Intuitively this is clear if you think that in order to make a mass oscillate with a given frequency and amplitude - and without any energy expenditure! - it is sufficient to add a single spring. And softness is robots is a generalization of a discrete spring.
Unfortunately, we are not there yet. Indeed, the community still lacks general techniques for describing, analysing, and exciting nonlinear oscillations in complex robotic systems.
To this end we recently introduced the concept of nonlinear modes of non-euclidean systems, which are discussed here in all their complexity (be aware that a much simpler - but less precise - formulation is available and used in follow up papers):
https://www.sciencedirect.com/science/article/pii/S1367578820300687
The aim of the activities within this topic is to extend tools and techniques that we are developing for articulated soft robots to continuum soft robots.