Recent robotic research recognized the advantages that Variable Impedance Actuators would yield to a new generation of robots, rendering them adapt to many different tasks of everyday life.

The importance of Variable Stiffness Actuators (VSA) in safety and performance of robots has been extensively discussed in the last decade. It has also been shown recently that a VSA brings performance advantages with respect to common actuators. For instance, the solution of the optimal control problem of maximizing the speed of a VSA for impact maximization at a given position with free final time is achieved by applying a control policy that synchronizes stiffness changes with link speed and acceleration. This problem can be regarded as the formalization of the performance of a soccer player's free kick.

The control of a robot's mechanical impedance is attracting increasing attention of the robotics community. Recent research in Robotics has recognized the importance of Variable Stiffness Actuators (VSA) in safety and performance of robots. An important step in using VSA for safety has been to understand the optimality principles that regulate the synchronized variation of stiffness and velocity when moving in the shortest time while limiting possible impact forces (the safe brachistochrone problem). In this paper, we follow a similar program of understanding the use of VSA in performance enhancement, looking at very dynamic tasks where impacts are maximized. To this purpose we address a new optimization problem that consists in choosing the inputs for maximizing the velocity of a link at a given final position, such as, e.g., for maximizing the effect of a hammer impact. We first study the problem with fixed stiffness, and show that, under realistic modeling assumptions, there does exist an optimal linear spring for the given inertia and motor. We then study optimal control of VSA and show that varying the spring stiffness during the execution of the hammering task improves the final performance substantially. The optimal control law is obtained analytically, thus providing insight in the optimality principles underpinning general VSA control. Finally, we show the practicality of our theoretical results with experimental tests.

This paper focuses on casting manipulation and presents an innovative mechanism that allows objects placed in a three-dimensional space to be reached with a relatively small robot. Casting manipulation is a technique by which a robotic end-effector is thrown and its ballistic flight is controlled through a tether cable so as to reach a target object. Previous work presented a solution that is viable only when the position of the target object is perfectly known or aligned with the throwing plane. Our work extends the technique by use of a novel mechanical design of the arm, and a suitable control scheme for the flight of the end-effector, which makes it applicable for objects placed at generic positions in a three-dimensional environment. Effectiveness of the casting robot was shown during the first ESA Lunar Robotics Challenge, where our team came in the second place.

In this paper we show that for an observer moving in the plane with no other information than the measurement of relative bearing to three known landmarks, it is possible to completely reconstruct its position and velocity. In particular this applies to the case where no model of the vehicle, nor odometry or acceleration measurements are available. Furthermore, in the same hypotheses, the position of any further landmark can be reconstructed from its bearing only. These results are more general than what is currently known on nonlinear observability of the SLAM problem, which relies on known observer velocities. Our results are also more general than the 2D version of known structure-from-motion observability results, which assume unknown but constant velocities. The proposed method is used to build a nonlinear observer directly applicable to a range of problems from computer vision to autonomous visual navigation.

This paper describes the design and implementation of DAVID, a lunar vehicle developed for the European Space Agency (ESA) Lunar Robotics Challenge, presenting severe terrain negotiation and sample acquisition challenges. We discuss in some detail two of the main innovative aspects of our entry to the challenge, i.e. the locomotion system and the sample acquisition system. Motivated by the challenge specifications, a range of different locomotion systems were considered, among which we chose a simple, rugged and effective wheeled system. We provide an account of the choice of five different types of wheels, which were designed, analyzed and experimentally tested in conditions similar to the challenge. The system eventually turned out to be very effective in negotiating 89% slopes of volcanic terrain on the challenge site, Mount Teide in Tenerife. To reduce the distance to be traveled on the difficult terrain and avoid risks in reaching the lowest parts of a crater, the vehicle was endowed with an innovative sample acquisition system, i.e. a casting manipulator. Casting manipulation is a technique in which the end-effector is thrown, the sample material is acquired, and the end-effector is retrieved using a light tether that acts as a "fishing line". The casting manipulator developed for DAVID uses an innovative sling-like technique, capable to obtain longer and more precise casts than previous oscillating versions. The analysis and experimental verification of DAVID's robot sling are reported, demonstrating its effectiveness. Finally, we give a brief account of the outcomes of the ESA Lunar Robotics Challenge, where our team came in second over other 8 teams that passed the final qualification phase.

This paper presents a correct solution to the optimal visual feedback control for a nonholonomic vehicle with limited field-of-view. Previous work on this subject has shown that the search for a shortest path can be limited to simple families of trajectories. We preliminarily provide an extension of the alphabet of optimal control words, to cover some regions of the vehicle plane where the synthesis of turns out to be suboptimal. The main contribution of this paper is an algorithm to translate the optimal synthesis to the image plane, thus enabling a purely image-based optimal control scheme. This allows better performance and increases the robustness of the overall process, avoiding the need of slowly-converging and error-prone parameter estimation algorithms. Simulations and experiments are reported which demonstrate the effectiveness of the proposed technique.