#	Pagina
attuale pagina	/open-h2020/projects/194280/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 2 - COMANOID (Multi-contact Collaborative Humanoids in Aircraft Manufacturing)

Teaser

Summary

The objectives of COMANOID is to research on deploying humanoid robots to achieve non-added value tasks for the workers that have been identified by Airbus Group in civilian airliner assembly operations. The project focuses on showing precise accessibility (namely into areas where wheeled robots cannot be deployed) through whole body multi-contact planning motion with advanced embedded 3D dense SLAM localization and visuo-force servoing capabilities. This will be demonstrated with a bracket positioning and printing or assembly task that does not require gripper manipulation dexterity. Because the robots evolve in human workers co-localized spaces, safety issues will be specifically accounted for. The results of COMANOID will be showcased in a 1:1 scale demonstrator of a real aircraft using two humanoid robots: the HRP-4 position controlled humanoid robot provided by CNRS partner and the TORO torque controlled humanoid robot provided by DLR partner. The technology that is developed is likely to apply to other large-scale manufacturing sites such as ship yards or even in construction and building at large.

Work performed

For M18, we achieved the following progress that is detailed in the technical document that accompany the first reporting period documents.

In WP1 we developed multi-contact retargeting methods in order to ease skill transfer from human workers to the humanoid robot in processing operation tasks. We also developed a posture generator method on the basis of non-linear optimization method and achieved some posture viability testing using HRP-4 and TORO and continued to work on possible new approaches for multi-contact planning and control. Visual servoing tasks are also developed and simulated on Airbus plane models using the HRP-4 and the QP controller. Visual servoing was also implemented in balancing and multi-contact strategies for balancing using new approaches were demonstrated on the TORO and on the HRP-4 humanoid robots. Humanoid walking algorithms are made much robust and can adapt on-line to perturbation and external forces. Selection of contact for additional support is integrated to the whole body control and a new approach for model preview control is proposed for robust walking. These are developed in simulation for the moment.

In WP2 we started to close the loop on the perception and the multi-contact planner. Indeed, premises of loop closing between real-time localization and mapping are experimented. We also investigated better registration approaches that have greater domain of convergence. It consists in researching how to best combine and fuse color and depth measurements for incremental pose estimation or 3D tracking. The idea is that most objects are known and if we can enhance the SLAM technique to (i) work well in the working site and (ii) to distinguish objects of interest, we could then easily close the loop on perception and investigate further refined planning and control strategies. We also started the work of extracting semantics from the COMANOID manufacturing environment. The environment can be provided by Airbus as a Digital MockUp (DMU) or acquired by the humanoid robot which maps continuously the evolving aircraft assembly line based on the results of localization and mapping generation. Last for the visual tracking, the first period focused on tracking the brackets provided by Airbus Group in order to grasp it and then place it on the aircraft fuselage at a given position. The first approach uses eye-in-hand camera and a model-based edge tracking.

In WP3, we started working on the development of a safety layer for humanoids and proposed a method for performing evasive motions with a humanoid robot. In the considered scenario, the robot is standing in a workspace, when a moving obstacle (e.g., a human, or another robot) enters its safety area and heads towards it; the humanoid must plan and execute in real-time a maneuver that avoids the collision. We also worked towards the extension of an existing framework with UNIROMA partner for human-robot safe coexistence and physical interaction and collaboration to encompass the COMANOID scenario. We have also started working on means to guarantee safety of humanoids and human co-workers through a combination of human aware robot planning, aimed at preventing the occurrence of risky situations, and purely reactive control strategies, triggered by the foreseen danger or actual occurrence of contact with humans. This will be continued by defining and planning integration of specific human-related safety criteria for predictability and legibility of motion as well as comfort of handovers. For the falling preventing and recovery we have developed various paths of research. We developed a balance control scheme which is able to decide online to use additional, optional support areas with the hands, when balance and safety are at stake. This is based on a generalization of the LMPC schemes developed for walking motions and has been demonstrated only in simulation so far. We also devised a method that considered falling by active postural reshaping and ad

Final results

The progress that is achieved in main COMANOID technologies is not ahead of the current state-of-the-art in multi-contact planning and control. We have achieved substantial advances in all key technologies as described in the technical document provided. We will be able to measure any impact on the socio-economical and societal tissues when the demonstrator will be achieved. Note that for walking, we are far from what Boston Dynamics is capable of achieving, this is due to the quality of their hardware, the funding discrepancy and the man-power.