• dual-arm
• self-collision avoidance
• teleoperation
• wearables

Teleoperation gives humans the possibility to act in hazardous or inaccessible environments, such as allowing remote care operations in hospital isolation wards reducing the risk of infection transmission. Those environments present different challenges for robotic manipulation, because they are complex, dynamic and uncontrolled. Bimanual manipulation plays a fundamental role in interacting with objects and enables to overcome some of these challenges. However, dual-arms systems introduce other issues related to self-collisions avoidance. Two types of self-collision are possible: collision of an arm with itself and collision of an arm with the other one.
The present master’s thesis aims to extend a previously developed single-arm teleoperation platform into a dual-arm one, by including a self-collision avoidance algorithm.
The teleoperated system developed comprises two 6-axis industrial robots equipped with two anthropomorphic artificial hands as end-effector. The operator’s motion is acquired thanks to a wearable system composed of an inertial motion capture suit and two sensorized gloves and then used to operate the teleoperated system. Concerning the management of the self-collision, the literature shows a valid solution limited to the case of collisions occurring between an arm with itself: CollisionIK, Rakita et al. (2021). It returns the robot joint angles by solving an IK optimization problem based on the target end-effector pose. However, this algorithm does not consider collisions between the two arms/hands.
To this aim, in this thesis we investigated to two different strategies, both based on the calculation of distances between right arm’s links and left’s ones but differing in the collision control mode and for complexity.
The first strategy monitors the output of CollisionIK and stops the teleoperated system if the distance between the parts of the robot overcome predefined thresholds. This simple strategy allows to safely operate with two arms/hands but led to forced interruptions of the manipulation if the arms/hands are too close.
The second strategy modifies the CollisionIK algorithm by feeding the optimization problem with a cost function that considers the links’ distances. This strategy allows a continuous operation of the robots by providing an alternative set of joint angles that avoid the collision. On the other hand, it increases the computational effort and may led to a pose of the robot slightly different respect to the actual pose of the operator.
An experiment with naïve teleoperators has been carried out to assess the self-collision avoidance strategies during the bimanual manipulation of objects with different lengths. The evaluation is based on objective metrics (completion time, number and type of error and accuracy) and subjective metrics (questionnaire). Overall, our results suggest that the second strategy drastically increases the number of successes attempts in manipulating short objects. Although it increases the error between the target and the executed trajectories, this does not affect the completion time. In addition, teleoperators prefer the second strategy since it allows to easily manipulate short objects without perceiving any difference in terms of delay and fidelity of movement reproduction.