Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - Co4Robots (Achieving Complex Collaborative Missions via Decentralized Control and Coordination of Interacting Robots)

Teaser

Imagine a future social/industrial facility where robots have been deployed to provide services and accomplish everyday tasks such as object handling/transportation, or pickup and delivery operations. In such a context, different robots with different actuation, manipulation...

Summary

Imagine a future social/industrial facility where robots have been deployed to provide services and accomplish everyday tasks such as object handling/transportation, or pickup and delivery operations. In such a context, different robots with different actuation, manipulation and perception capabilities must be coordinated in order to achieve various complex tasks that require collaborative actions with each other and with humans. Thus, the supervision and coordination of the overall heterogeneous system mandates a decentralized framework that integrates high-level task-planning, low-level motion control and robust, real-time robot perception. The World Robotics Survey points out that indoor logistics become a major market. Sales of professional service robots in the logistics segment registered a growth of 28% in 2014 with the value of sales increasing to USD 2.2 billion. This shows the opportunity for European companies to address this area. Current practice in coordination of robotic teams is mostly based on offline, centralized planning and related tasks are almost exclusively fulfilled in a predefined manner. Only a little room is allowed for real-time and coordinated decentralized actions. We argue that this does not utilize the capabilities of multi-robot systems to operate efficiently in a dynamic environment. In most cases, sudden changes in the environment, the type of assigned tasks, and the need for coordination, would cause the system to halt, ask for a human intervention and restart. Thus the need for a framework for decentralized, real-time, automated task (re)-planning is evident. Based on the above, within Co4Robots our goal is to build i. a systematic real-time decentralized methodology to accomplish complex mission specifications given to a team of potentially heterogeneous robots, ii.a set of control schemes appropriate for the mobility and manipulation capabilities of the considered robotic platforms, their types and dynamics, the unexpected and sudden changes in the environment, as well as the presence of humans, iii. a set of perceptual capabilities that enable robots to localize themselves and estimate the state of their highly dynamic environment, in the presence of strong interactions and in a collaborative manner, and iv., their corresponding systematic integration at both the conceptual and the software implementation levels.

Work performed

During the period, a set of strategies for manipulation and mobility have been developed. We tackled Load Exchange among heterogeneous agents, with a vision-based grasping and motion control scheme, a planning algorithm, ands an optimal grasping selection scheme. Regarding Cooperative Load Transportation, the solutions provided so far include human-robot collaboration in obstacle-cluttered environments via implicit communication, as well as a compliance control for human–robot cooperative manipulation. Regarding Cooperative Motion Planning and Collision Avoidance, the contributions so far include navigation algorithms, as well as decentralized and reconfigurable multi-agent control schemes. A decentralized real-time planning framework for a team of heterogeneous agents was also designed. The input is a complex task given as a linear temporal logic (LTL) or metric interval temporal logic (MITL) formula. A centralized authority may assign a task to agents, while also local tasks are permitted. The automated plan synthesis operates in three steps. First, efficient decentralized methods translate each agents\' motion capabilities and dynamic interaction into a discrete representation. Second, the initial task is decomposed into local ones. Third, a high-level plan is obtained. In addition, robot perception algorithms have been developed, resulting in several publications in the fields of human and hand pose estimation, and action recognition. Furthermore, algorithms that enable the robot to detect and track objects and humans, and recognize human gestures have been integrated into the project’s robotic platform. Finally, the software architecture (SERA) for the project has been developed. SERA is an architecture for robot applications that supports human-robot collaboration, as well as adaptation and coordination of single- and multi-robot systems in a decentralized fashion. SERA facilitates the process of developing and deploying software into a robot. In addition, we developed a Domain Specific Language, which aims at providing an alternative imperative language that defines robot tasks.

Final results

Although the most employed communication form in multi-agent systems is the explicit one, in case of faulty communication, severe problems may arise, whereas increasing the number of robots will request complex protocols to deal with crowded bandwidth. Several of the aforementioned limitations can be overcome by employing implicit communication instead. Despite the increased difficulty of the analysis, it leads to simpler protocols and saves bandwidth as well as power. Our goal is to design an autonomous system consisting of multiple heterogeneous robots, aiming at solving tasks using interactions among them, both explicitly at a symbolic level and implicitly at the low level. The challenge will lie in replacing heavy explicit communication with implicit, that results indirectly from the robot interaction. Moreover, although LTL can capture many temporal tasks, it does not support explicit time bounds. One of the goals of Co4Robots is to develop techniques for automated synthesis under local MITL tasks that involve quantitative time bounds, starting from individual specifications for each agent in the form of MITL. A novelty here is the assignment of specifications to the team in a manner of partial decentralization. Moreover, we plan to parametrize the degree of decentralization and increase it accordingly when tasks become infeasible or do not fulfill a desired satisfaction robustness threshold. The goal of perception in the context of Co4Robots is to aid interaction and collaboration among robots/humans.This can enhance the performance of perception algorithms such as SLAM and object or agent detection and tracking. Our goal is twofold: to increase the perception abilities of each individual robotic platform and allow for enhanced overall perception of the environment through robots and humans in collaboration. To increase the perception abilities and autonomy of each individual robot we plan on bringing together various vision technologies close to the sensors. This includes a) the combination of FORTH’s SLAM, 3D tracking framework and additionally object detection so that they effectively cooperate on a single computational platform in the embodiment of the robot and b) their modification so as to enable real-time processing rates. Software engineering is called to play a key role in securing the new technology’s affirmation by making it pervasive and ubiquitous. We plan to use Model-driven Engineering (MDE) techniques. MDE has been identified as key technology that has reached mature level in other domains, but not in robotics. In fact, the use of models represents an important step towards easy portability and re-deployability. This calls for platform neutrality. The ideal way to reach this goal is the standardization of basic systems functions, scalability to different robots and platform variants, maintainability throughout the robot life-cycle and software updates over the robot’s lifetime.

Website & more info

More info: http://www.co4robots.eu/.