We present an avatar system that enables a human operator to visit a remote location via iCub3, a new humanoid robot developed at the Italian Institute of Technology (IIT) paving the way for the next generation of the iCub platforms. On the one hand, we present the humanoid iCub3 that plays the role of the robotic avatar. Particular attention is paid to the differences between iCub3 and the classical iCub humanoid robot. On the other hand, we present the set of technologies of the avatar system at the operator side. They are mainly composed of iFeel, namely, IIT lightweight non-invasive wearable devices for motion tracking and haptic feedback, and of non-IIT technologies designed for virtual reality ecosystems. Finally, we show the effectiveness of the avatar system by describing a demonstration involving a realtime teleoperation of the iCub3. The robot is located in Venice, Biennale di Venezia, while the human operator is at more than 290km distance and located in Genoa, IIT. Using a standard fiber optic internet connection, the avatar system transports the operator locomotion, manipulation, voice, and face expressions to the iCub3 with visual, auditory, haptic and touch feedback.

Balancing and push-recovery are essential capabilities enabling humanoid robots to solve complex locomotion tasks. In this context, classical control systems tend to be based on simplified physical models and hard-coded strategies. Although successful in specific scenarios, this approach requires demanding tuning of parameters and switching logic between specifically-designed controllers for handling more general perturbations. We apply model-free Deep Reinforcement Learning for training a general and robust humanoid push-recovery policy in a simulation environment. Our method targets high-dimensional whole-body humanoid control and is validated on the iCub humanoid. Reward components incorporating expert knowledge on humanoid control enable fast learning of several robust behaviors by the same policy, spanning the entire body. We validate our method with extensive quantitative analyses in simulation, including out-of-sample tasks which demonstrate policy robustness and generalization, both key requirements towards real-world robot deployment.

Human-like trajectory generation and footstep planning represent challenging problems in humanoid robotics. Recently, research in computer graphics investigated machinelearning methods for character animation based on training human-like models directly on motion capture data. Such methods proved effective in virtual environments, mainly focusing on trajectory visualization. This paper presents ADHERENT, a system architecture integrating machine-learning methods used in computer graphics with whole-body control methods employed in robotics to generate and stabilize human-like trajectories for humanoid robots. Leveraging human motion capture locomotion data, ADHERENT yields a general footstep planner, including forward, sideways, and backward walking trajectories that blend smoothly from one to another. Furthermore, at the joint configuration level, ADHERENT computes data-driven wholebody postural reference trajectories coherent with the generated footsteps, thus increasing the human likeness of the resulting robot motion. Extensive validations of the proposed architecture are presented with both simulations and real experiments on the iCub humanoid robot, thus demonstrating ADHERENT to be robust to varying step sizes and walking speeds.