Figure 1: Bert is an articulated soft quadruped, designed to walk efficiently and robustly by copying the natural example. Most of the proposed theses will find application in control algorithm implemented on this systems.
This is only a partial list of available topics. If you are interested in robotics locomotion, flexible and soft robotics, nonlinear dynamics and control, and you cannot find any theses that you like in the following list, you may want to send us an email anyway, to see if there are other opportunities. Please, also take a look at https://www.in.tum.de/i23/available-theses/
Dynamic locomotion algorithms for soft quadrupeds
Locomotion of legged robots is a challenging problem due to its hybrid dynamics (discrete contact sequencing and continuous whole-body motion), and the constraints on the direction and amplitude of the contact forces. Recently, the concepts of three-dimensional Divergent Component of Motion (DCM) and Virtual Repellent Point (VRP) were introduced in [1], decomposing the second-order CoM dynamics into two first-order linear dynamics, with the CoM converging to the DCM (stable dynamics), and the DCM diverging away from the VRP (unstable dynamics). Based on this formulation, continuous closed-form DCM and CoM trajectories can be generated using a piecewise interpolation of the VRP trajectory over a sequence of waypoints [2]. This highly compact motion representation is a natural way of handling the locomotion hybrid dynamics, with the discrete contact sequencing being mapped onto the VRP waypoints. This approach has been used successfully for bipedal locomotion in [1], and dynamic multi-contact motion in [3].
This thesis will be about generalizing this concept to quadrupedal locomotion, and to understand to which extent this technique can be realized with intermitted control actions. The ultimate goal of this thesis will be the implementation of dynamic locomotion with the robot Bert [4].
Suggested as (but not limited to!)
References
Supervisors: George Mesesan (George.Mesesan[ @ ]dlr.de), Dr. Cosimo Della Santina (cosimodellasantina[ @ ]gmail.com), Prof. Dr.-Ing. Alin Albu-Schaeffer
Multiple theses in: Discovering natural oscillations in highly coupled mechanical systems with machine learning and data-driven strategies, with application to soft robotics.
For the study of highly nonlinear mechanical systems, finding special periodic solutions which are generalization of the well-known normal modes of linear systems promise to enable the execution of complex tasks (e.g. running, swimming, periodic pick and place) with an efficiency and a robustness close to the one observed in animals. However, the study of nonlinear normal modes is rather a niche topic, treated mainly in the context of structural mechanics for systems with Euclidean metrics, i.e., for point masses connected by nonlinear springs. Nonetheless newest results emphasize that a very rich structure of periodic and low-dimensional solutions exist also within nonlinear systems such as elastic multi-body systems encountered in the biomechanics of humans and animals or of humanoid and quadruped robots, which are characterized by a non-constant metric tensor. Evaluating an analytic mathematical description of nonlinear normal modes for these systems is, however, no simple task. The most straightforward way to do that is to approximate the solution of a set of partial differential equations describing the geometry of these structures, using Galerkin methods. Unfortunately, this method provides only local solutions, and do not scale well with the dimension of the dynamical system. As an alternative, this thesis will investigate the use of simulation based and data driven strategies as swarm and particle optimization, Bayesian optimization, (deep) autoencoders, and other machine learning techniques. Whenever possible, these techniques will be guided by the theoretical knowledge of the underlying mathematical structure, allowing for effective and efficient formulations of the problem.
The expected outcome of this thesis will be a semi-professional MatLab toolbox, able to discover nonlinear normal modes in generic soft robots. Therefore, either a good knowledge of this programming language, or above average programming skills are mandatory.
Supervisor: Prof. Dr.-Ing. Alin Albu-Schaeffer, Dr. Cosimo Della Santina (cosimodellasantina[ @ ]gmail.com)
Deterministic chaos, resonant motions, and other nonlinear behaviours in a generalization of the elastic pendulum
Controlling motion at low energetic cost, both from mechanical and computational point of view, certainly constitutes one of the major locomotion challenges in biology and robotics. Recent results suggest that robots can be designed and controlled to move efficiently by exploiting resonance body effects, increasing the performance compared to rigid body designs. This motivates the study of simple yet meaningful mechanical syestems, which represent complex behaviours through few degrees of freedoms (e.g. SLIP model for legged locomotion). To such end, this thesis will focus on analysing a generalization of the elastic pendulum, which includes a fully coupled elastic field rather than a solely radial one. This system can be seen as a paradigm for an elastic leg. Of particular interest will be the study of complex nonlinear behaviours (e.g. chaos, instabilities, bifurcations) arising when either mechanical parameters (e.g. body inertia, spring stiffness), or energy levels change.
The workload will be modulated on the curricular needs of the applicant (e.g. master thesis longer than bachelor thesis).
Supervisors: Prof. Dr.-Ing. Alin Albu-Schaeffer, Dr. Cosimo Della Santina (cosimodellasantina[ @ ]gmail.com)