Quadrupedal locomotion skills are challenging to develop. In recent years, deep Reinforcement Learning promises to automate the development of locomotion controllers and map sensory observations to low-level actions. Moreover, the full robot dynamics model can be exploited, but no model-based simplifications are to be made. In this work, a method for developing controllers for the Laelaps II robot is presented and applied to motions on slopes up to 15°. Combining deep reinforcement learning with trajectory planning at the toe level, reduces complexity and training time. The proposed control scheme is extensively tested in a Gazebo environment similar to the treadmill-robot environment at the Control Systems Lab of NTUA. The learned policies produced promising results.

This paper studies the effects of varying rollover curvature on the passive dynamic gait of a biped walker. The dynamic model of a compliant biped robot is extended with the implementation of semielliptical feet, to mimic human rolling-radius progression during a step. The process of modeling the semielliptical foot shape and integrating its kinematics to the biped's dynamics is presented in detail. The passive dynamic behavior of the biped for elliptic feet of various dimensions is investigated through numerical simulations to provide results about gait stability, walking speed, energetic efficiency, and impact force levels. The concept of energetic efficiency in passive walking is discussed thoroughly, and an efficiency comparison methodology is proposed. Finally, it is shown that the biomimetically-inspired semielliptical foot profile can lead to higher gait efficiency. The results of this study can be used to optimize energetic efficiency in biped walking machines and/or gait assisting prosthetic equipment by means of foot shape optimization.

Drawing on screw theory and the virtual joint method, this paper presents a general and hierarchical approach for semianalytical stiffness modeling of parallel mechanisms. The stiffness model is built by two essential steps: 1) formulating the map between the stiffness matrices of platform and limbs using the duality of wrench and twist of the platform; and 2) formulating the map between stiffness matrices of a limb and a number of elastic elements in that limb using the duality of the wrench attributed to the limb and the twist of the endlink of that limb. By merging these two threads, the Cartesian stiffness matrix can be explicitly expressed in terms of the compliance matrices of joints and links. The proposed approach bridges the gap between two currently available approaches and is thereby very useful for evaluating stiffness over the entire workspace and investigating the influences of joint/link compliances on those of the platform in a quick and precise manner. A stiffness analysis for a 3-PRS parallel mechanism is presented as an example to illustrate the effectiveness of the proposed approach.

The notions of manipulator layout, layout center, layout length, and layout conditioning for any serial-type robotic manipulators are discussed. The motivation behind these concepts is the devising of a kinetostatic performance index to serve in the optimum dimensioning of the geometric parameters of serial manipulators. Based on these definitions, a measure of manipulator conditioning is devised that is highly suited for the intended tasks of manipulator design. Moreover, this performance index is substantially less expensive to compute than other measures of kinematic conditioning.

The paper discusses extensions to the classical transmission-element paradigm which are suitable for the efficient modeling of mechanical systems. The modeling is based on three characteristic mechanical features (1) motion transmission, including position, velocity, and acceleration, (2) force feedback, and (3) propagation of inertia. These aspects can be given a differential-geometric interpretation, making the approach suitable for modeling tasks in more general settings, as in mechatronic systems.