Cooperative lifting (co-lift) is an important application of HRI with use-cases in many fields such as manufacturing, assembly, medical rehabilitation, etc. Successful industrial implementation of co-lifting requires the operations of approaching, attaching, lifting, carrying and placing the object to be handled as a whole rather than individually. In this paper, we target all stages of cooperative lifting in a holistic approach and extend previous results in [1] using IMU-based human motions estimates by introducing force-based control. We demonstrate through experiments on a UR5e robot how the force-based approach significantly improves on the position-based approach of [1]. Additionally, we improve the real-time control capabilities of the system by using a real-time data exchange communication interface. We believe that our system can be an advancing point for more human motion/gesture-based HRI applications as well as increasing the uptake of human-robot co-lifting systems in industrial settings.

For a successful deployment of physical Human-Robot Cooperation (pHRC), humans need to be able to teach robots new motor skills quickly. Probabilistic movement primitives (ProMPs) are a promising method to encode a robot’s motor skills learned from human demonstrations in pHRC settings. However, most algorithms to learn ProMPs from human demonstrations operate in batch mode, which is not ideal in pHRC when we want humans and robots to work together from even the first demonstration. In this paper, we propose a new learning algorithm to learn ProMPs incre-mentally and continuously in pHRC settings. Our algorithm incorporates new demonstrations sequentially as they arrive, allowing humans to observe the robot’s learning progress and incrementally shape the robot’s motor skill. A built-in forgetting factor allows for corrective demonstrations resulting from the human’s learning curve or changes in task constraints. We compare the performance of our algorithm to existing batch ProMP algorithms on reference data generated from a pick-and-place task at our lab. Furthermore, we demonstrate how the forgetting factor allows us to adapt to changes in the task. The incremental learning algorithm presented in this paper has the potential to lead to a more intuitive learning progress and to establish a successful cooperation between human and robot faster than training in batch mode.

Two different configurations of Liquid State Machine (LSM), a special type of Reservoir Computing with internal nodes modelled as spiking neurons, implementing multiple columns (Modular and Monolithic approaches) are tested against the decimation of neurons, connections and entire columns in order to verify which one can better withstand the damage. Based on the neurorobotics outlook, this work is part of a bigger project that aims to apply artificial neural networks to the control of humanoid robots. Therefore, as a benchmark, we made use of a robotic task where an LSM is trained to generate the joint angles needed to command a simulated version of the collaborative robot BAXTER to draw a square on top of a table. The final drawn shape is analysed through Dynamical Time Warping to generate a cost value based on how close the produced drawing is to the original shape. Our results show both approaches, Modular and Monolithic, had a similar behaviour, however the Modular was better at withstanding the decimation of neurons when it was concentrated in a single column.

Biological neural networks are able to control limbs in different scenarios, with high precision and robustness. As neural networks in living beings communicate through spikes, modern neuromorphic systems try to mimic them making use of spike-based neuron models. Liquid State Machines (LSM), a special type of Reservoir Computing system made of spiking units, when it was first introduced, had plasticity on an external layer and also through Short-Term Plasticity (STP) within the reservoir itself. However, most neuromorphic hardware currently available does not implement both Short-Term Depression and Facilitation and some of them don't support STP at all. In this work, we test the impact of STP in an experimental way using a 2 degrees of freedom simulated robotic arm controlled by an LSM. Four trajectories are learned and their reproduction analysed with Dynamic Time Warping accumulated cost as the benchmark. The results from two different set-ups showed the use of STP in the reservoir was useful for one out of three tested trajectories, though not computationally cost-effective for this particular robotic task.

The work described here explores a neural network architecture that can be embedded directly in the realtime sensorimotor coordination loop of a developmental robot platform. We take inspiration from the way children are able to learn while interacting with a teacher, in particular the use of prediction of the teacher actions to improve own learning. The architecture is based on two neural networks that operate online, and in parallel, one for learning and one for prediction. A Hebbian learning rule is used to associate the high-dimensional afferent sensor input at different time-delays with the current efferent motor commands corresponding to the teacher demonstration. The predictions of future motor commands are used to limit the growth of the neural network weights, and to enable the robot to smoothly continue movements the teacher has begun. Results on a simulated iCub robot learning object interaction tasks are presented, including an analysis of the sensitivity to changes in the task setup. We also outline the first implementation on the real iCub platform.

An approach for adaptive shared control of an assistive manipulator is presented. A set of distributed collision and proximity sensors is used to aid in limiting collisions during direct control by the disabled user. Artificial neural networks adapt the use of the proximity sensors online, which limits movements in the direction of an obstacle before a collision occurs. The system learns by associating the different proximity sensors to the collision sensors where collisions are detected. This enables the user and the robot to adapt simultaneously and in real-time, with the objective of converging on a usage of the proximity sensors that increases performance for a given user, robot implementation and task-set. The system was tested in a controlled setting with a simulated 5 DOF assistive manipulator and showed promising reductions in the mean time on simplified manipulation tasks. It extends earlier work by showing that the approach can be applied to full multi-link manipulators.

Assistive robots are increasingly being envisioned as an aid to the elderly and disabled. However controlling a robotic system with a potentially large amount of Degrees of Freedom (DOF) in a safe and reliable way is not an easy task, even without limitations in the mobility of the upper extremities. Shared control has been proposed as a way of aiding disabled users in controlling mobility aids such as assistive wheelchairs, by using the sensors of the robotic platform to predict the user's intent and assist in navigation. Assistive manipulators, that aim to perform physical Daily Life Activities (DLA), is a more complex problem however. This calls for good experimental practices to ensure repeatability, reproducibility, and steady progress. The work presented here attempts to model the complete system for assistive manipulators, and in the context of this model define metrics and good practices for benchmarking shared control for such robots. An adaptive shared control approach for limiting collisions during teleoperation is used as a case study. Improvements in performance are shown, quantified by the trade-off between mean time and number of collisions as well as the controllability from the user's perspective.

Assistive robotics HRI has a number of important characteristics that distinguishes it from other forms of HRI. This includes the need for both high flexibility, safety and reliability in controlling the robotic system. Approaching the system as a human-robot binomial, with the user and the robot acting in a closed-loop, may be beneficial to understanding and improving the interaction. This paper investigates the feasibility of modeling and quantifying assistive robotics HRI inside such a human-robot binomial using concepts from Information Theory.

ASIBOT is an assistive climbing robot that is capable of aiding in daily tasks from fixed docking stations in the environment. A task-oriented design process was applied to improve the robot kinematic structure, which was based on the grid method. Twelve different robot designs were optimized for typical kitchen scenarios, followed by a quantitative comparison.