Ricerc@Sapienza

We present a novel framework for the automatic discovery and recognition of motion primitives in videos of human activities. Given the 3D pose of a human in a video, human motion primitives are discovered by optimizing the ‘motion flux’, a quantity which captures the motion variation of a group of skeletal joints. A normalization of the primitives is proposed in order to make them invariant with respect to a subject anatomical variations and data sampling rate.

In recent years Reinforcement Learning (RL) has achieved remarkable results. Nonetheless RL algorithms prove to be unsuccessful in robotics applications where constraints satisfaction is involved, e.g. for safety. In this work we propose a control algorithm that allows to enforce constraints over a learned control policy. Hence we combine Nonlinear Model Predictive Control (NMPC) with control-state trajectories generated from the learned policy at each time step. We prove the effectiveness of our method on the Pendubot, a challenging underactuated robot.

We highlight the equivalence between the motion of an elastic joint and the two-body problem in classical mechanics. Based on this observation, a change of coordinates is introduced that reduces the two-body problem to a pair of decoupled one-body problems. This allows to treat the rest-to-rest motion problem with bounded actuator torque in an elegant geometric fashion. Instead of dealing directly with the fourth-order dynamics, we consider two equivalent masses whose motions have to be synchronized in separate phase spaces.

Introduction: Robot-assisted therapy is an emerging approach that performs highly repetitive, intensive, task oriented and quantifiable neuro-rehabilitation. In the last decades, it has been increasingly used in a wide range of neurological central nervous system conditions implying an upper limb paresis. Results from the studies are controversial, for the many types of robots and their features often not accompanied by specific clinical indications about the target functions, fundamental for the individualized neurorehabilitation program.

In recent years, the evolution of artificial intelligence techniques has widely grown such that it gives new ways to improve human life, not only at work but also living. Nowadays, to the human being, physical human-robot interactions (PHRIs) have been presented very important and present itself as a major challenge for the current engineering. Therefore, this work designs and analyses a two-degree-of-freedom robotic arm with flexible joints driven by a DC motor.

A linear-quadratic optimal control problem is considered for the infinite-dimensional model of a one-link flexible arm. Two boundary inputs are assumed to be available, namely the joint torque at the link base and a transverse force at the tip of the link. The problem is formulated and solved using semigroup theory and duality arguments. Simulation results are provided to support the theoretical findings, comparing the proposed optimal LQ law with a more conventional PD/state feedback controller in terms of cost and transient performance.

We report on the results of a collaborative project that investigated the deployment of humanoid robotic solutions in air-craft manufacturing for several assembly op erations where access by wheeled or railported robotic platforms is not possible.

Over the years, there have been many improvements in job-related safety standards and working conditions, but there are still many situations and environments where human lives are put at risk, such as in search and rescue situations, construction sites, and chemical plants. We envision a world where robots can act as physical avatars and effectively replace humans in those hazardous scenarios through teleoperation.

We consider a situation in which a humanoid robot must reach a goal region (walk-to task) walking in an environment consisting of horizontal patches located at different heights (world of stairs). To solve this problem, the paper proposes an integrated motion planner/controller working in two stages: off-line footstep planning and on-line gait generation. The planning stage is based on a randomized algorithm that efficiently searches for a feasible footstep sequence.

A control system based on multiple sensors is proposed for the safe collaboration of a robot with a human. New constrained and contactless human-robot coordinated motion tasks are defined to control the robot end-effector so as to maintain a desired relative position to the human head while pointing at it. Simultaneously, the robot avoids any collision with the operator and with nearby static or dynamic obstacles, based on distance compu- tations performed in the depth space of a RGB-D sensor.