Fabio DallaLibera

Design and control of a ray-mimicking soft robot based on morphological features for adaptive deformation

Underwater tasks are diversified and articulated. The environment in which they must be accomplished is often unconstrained and unpredictable. Operating AUVs assuring safety of the robot and of its surrounding is therefore very difficult. On the other hand, many fishes are able to easily move in the same environments. A crucial factor for this capability is their body, which consists primarily of elastic and soft structures that enable both complex movement and adaptation to the environment. Among the most efficient swimmers we find rays, which show abilities like high speed turning and omnidirectional swimming. In this paper we propose an underwater soft robot based on the morphological features of rays. We mimic both their radially skeletal structure with independent actuators for each bone and the compliance of their fins. This flexibility of the structure provides an adaptive deformation that allows our robot to swim smoothly and safely.

How do People Expect Humanoids to Respond to Touch

With close interaction between humans and robots expected to become more and more frequent in the near future, tactile interaction is receiving increasing interest. Many advances were made in the fields of tactile sensing and touch classification. Robot’s reactions to touches are usually decided by the robot’s designers and fit to a particular purpose. However, very little investigation has been directed to the movements that common people expect from robots being touched. This paper provides an initial step in this direction. Responses that people expect from a humanoid being touched were collected. These responses were then classified by automatically grouping similar responses. This allows the identification of distinct types of responses. Evaluation of how this grouping matches common sense were then performed. Results showed strong correlation between the automatic grouping and common sense, providing support to the idea that the automatically identified types of responses correspond to a plausible classification of robot’s responses to touch.

Biologically inspired mobile robot control robust to hardware failures and sensor noise

Some bacteria present a movement which can be modeled as a biased random walk. Biased random walk can be used also for artificial creatures as a very simple and robust control policy for tasks like goal reaching. In this paper, we show how a very simple control law based on random walk is able to guide mobile robot equipped with an omnidirectional camera toward a target without any knowledge about the robots actuators or about the robots camera parameters. We verified, by several simulation experiments, the robustness of the random biased control law with respect to failures of robots actuators or sensor damages. These damages are similar to the ones which can occur during a RoboCup match. The tests show that the optimal behavior is obtained using a bias which is roughly proportional to the random walk step, with a coefficient dependent on the physical structure of the robot, on its actuators and on and its sensors after the damage. Finally, we validated the proposed approach with experiments in the real world with a wheeled robot performing a goal reaching task in a Middle-Size RoboCup field without any prior knowledge on the actuators and without any calibration of the very noisy omnidirectional camera mounted on the robot.

A parameterless biologically inspired control algorithm robust to nonlinearities, dead-times and low-pass filtering effects

A biologically inspired control algorithm for robot control was introduced in a previous work. The algorithm is robust to noisy sensor information and hardware failures. In this paper a new version of the algorithm is presented. The new version is able to cope with highly non-linear systems and presents an improved robustness to low-pass filter effects and dead-times. Automatic tuning of the parameters is also introduced, providing a completely parameterless algorithm.

On the human perception of dissimilarities between postures of humanoids

Graphical Abstract This paper analyzes the perceived dissimilarity between postures of humanoid robots. First, the human perception of absolute distance between postures is focused. It is shown that results derived in computer graphics for human figures can be replicated with humanoids, despite the difference in body proportions and arrangement of the degrees of freedom. Successively, the perception of relative distances between postures is considered. It is shown that, paradoxically, better prediction of the distance between a pair of postures does not necessary lead to better predictions of which of multiple distances is the shortest. Finally, the paper concludes by briefly discussing possible implications of this finding to the fields of motion retrieval and motion blending.

Stochastic resonance emergence from a minimalistic behavioral rule

Stochastic resonance (SR) is a phenomenon occurring in nonlinear systems by which the ability to process information, for instance the detection of weak signals is statistically enhanced by a non-zero level of noise. SR effects have been observed in a great variety of systems, comprising electronic circuits, optical devices, chemical reactions and neurons. In this paper we report the SR phenomena occurring in the execution of an extremely simple behavioral rule inspired from bacteria chemotaxis. The phenomena are quantitatively analyzed by using Markov chain models and Monte Carlo simulations.

Humanoid motion representation by sensory state transitions

Given the complexity and the high number of degrees of freedom humanoid robot motions often are generated off-line, stored in the robot memory and then reproduced. In the simplest cases motor commands are just replayed in open-loop (i.e. feed-forward), in more advanced implementations few parameters are changed on fly to adapt the motion to the external conditions. Indeed, several authors proposed a large variety of techniques to exploit the input from sensory feedback to modify a reference trajectory, in order to cope with environment changes and disturbances. In this paper, we propose a motion representation for humanoid robots that includes the sensory feedback information in the motion representation itself. This motion description allows stabilization against disturbances and environmental changes, but does not require any design or tuning of the relationships between sensory inputs and movement modification. We present experimental results on a simulated small humanoid robot equipped with motor encoders and touch sensors covering the whole body.

Noise-modulated neural networks as an application of stochastic resonance

Stochastic resonance (SR) is a phenomenon by which the input signal of a nonlinear system, with magnitude too small to affect the output, becomes observable by adding a non-zero level of noise to the system. SR is known to assist biological beings in coping with noisy environments, providing sophisticated information processing and adaptive behaviors. The SR effect can be interpreted as a decrease in the input-output information loss of a nonlinear system by making it stochastically closer to a linear system. This work shows how SR can improve the performance of a system even when the desired input-output relationship is nonlinear, specifically for the case of a neural networks whose hidden layers consist of threshold functions. Universal approximation capability of neural networks exploiting SR is then discussed: although a network consisting of threshold activation functions has been proven to be an universal approximator in the context of the extreme learning machine (ELM), once SR is taken into account, the system can be deemed as a classic three-layer neural network whose universality has been previously proven by simpler proofs. After proving the universal approximation capability for an infinite number of hidden units, the performance achieved with a finite number of hidden units is evaluated using two training algorithms, namely backpropagation and ELM. Results highlight the SR effect occurring in the proposed system, and the relationship among the number of hidden units, noise intensity, and approximation performance.

Kinematic Analysis of a 3D Printable 4-DOF Desktop Robot Actuated Exclusively by Revolute Pairs

This paper describes the kinematic structure of Yondy, a desktop robot with three rotational DOFs and one translational DOF. The kinematic structure comprises only rotational joints, easing its construction using off-the-shelf rotational actuators like servomotors. No pinion-rack or other mechanical elements are required for its construction, permitting its realization even with low-accuracy 3D printers. The robot is realized as a hybrid 4-DOF mechanism, with a planar 2-DOF parallel manipulator connected in series with other 2 DOFs. Particular focus is given to the parallel manipulator, which is interesting from a theoretical point of view because it is able to undergo nonsingular assembly mode transitions.

Control of real-world complex robots using a biologically inspired algorithm

Elementary living beings, like bacteria, are able to reach food sources using only limited and very noisy sensory information. In this paper, we describe a very simple algorithm inspired from bacteria chemotaxis. We present a Markov chain model for studying the effect of noise on the behavior of an agent that moves according to this algorithm, and we show that, counterintuitively, the application of noise can increase the expected average performance over a fixed available time. After this theoretical analysis, experiments on real-world application of this algorithm are introduced. In particular, we show that the algorithm is able to control a complex robot arm, actuated by 17 McKibben pneumatic artificial muscles, without the need of any model of the robot or of its environment.