Development of intelligent systems distributed with the robot operating system

0

Networked robotic systems are increasingly popular. In addition to industrial robotics now well entrenched in manufacturing, applications are also developing in logistics, medical technology and even unmanned aerial vehicles (UAVs) and unmanned underwater vehicles (UUVs).

In general, robotic systems are always centrally controlled by an API based on statically scheduled tasks. This means that the individual systems have a low level of autonomy and that the whole system shuts down if the central controller fails.

However, if these systems are intended to operate in an unfamiliar environment or need to collaborate with each other in a contextual manner, complex algorithms must be implemented, ideally directly in the robot’s control system itself. This allows a robot to respond with low latency, regardless of the reliability of its communication with a central controller.

Since these systems are generally safety critical, breakdowns and failures typically result in high financial damage or even dangerous situations for operating personnel in the case of robots designed to collaborate with humans.

To guard against such situations, it is necessary to consider the interactions and couplings between individual systems or between systems and the operator. These can be physical in nature, such as for collaboration on moving heavy and bulky loads, or the coupling between systems can be a communication system (usually wireless).

Scenarios leading to undesirable behavior can become so complex that safe operation is only made possible through methodical model-based development, as is already the case with mechatronic systems in aviation and space travel. as well as in the automotive industry.

Such a process is made possible by the Robot Operating System (ROS).

ROS provides an environment for the integration of many functions required by autonomous systems, such as sensors, actuators, navigation and path planning.

The community of developers of this open source solution provides a large number of complete modules for the rapid and systematic development of robotic systems. Major robotics manufacturers now also offer ROS modules for system development [1].

As middleware, ROS enables protocol-based data exchange between processes (in ROS: nodes), whether these are performed at a virtual level (as in a simulation) or linked to physical hardware. Implementation as a physically distributed system is also supported, in which case ROS runs in parallel on multiple computing platforms.

This enables end-to-end, model-based module development, such as sensor data analysis or actuator control. ROS also offers the option of implementing the corresponding algorithms in various programming languages, such as Python for quick and easy prototyping or C ++ for high performance and low latency. In addition to this, ROS provides interfaces to environments such as Gazebo ( http://gazebosim.org/ ), which simulates the kinematics of robotic systems and their environment. With the Functional Mock-up Interface (FMI) standard, it is also possible to integrate simulations of multiphysics systems with a variety of established tools, such as Modelica.

New versions of ROS can also be operated on on-board hardware with defined real-time capacity requirements. This allows nodes that were first developed in simulation and virtually validated to be ported directly to a microelectronic platform suitable for production use, such as a microcontroller with FPGA accelerator. [2].

Such hardware / software systems can be effectively validated with hardware-in-the-loop environments. These can be easily derived from a full system simulation via the ROS architecture, whether for an individual robotic system or for collaborative systems. To do this, ROS can establish communication between nodes running on robot control processors and other nodes that simulate actuators, sensors and the environment.

End-to-end orderly development processes for complex distributed systems can be implemented in this way. For example: learning systems of cyber-physical agents that physically interact with each other and also exchange information in order to achieve distributed methods, such as federated learning, in addition to centralized training methods such as learning by reinforcement.

Mobile systems in particular use wireless communication for these purposes. Even with modern broadband wireless standards, this requires consideration of the special properties of protocol-based wireless communication. These include some uncertainty regarding the transmission speed (latency) and possible communication faults resulting from the continuous evolution of the characteristics of the network. Such effects should be taken into account as early as possible during the system design process. Current work therefore focuses on the question of how the characteristics and imperfections of such communication systems can be integrated into the development of the system. [3].

The references

[1] Zhang, L., Merrifield, R., Deguet, A., & Yang, G.-Z. (2017). Powering the World’s Robots — 10 Years of ROS. Scientific robotics, 2 (11). https://doi.org/10.1126/scirobotics.aar1868

[2] Moréac, E., Abdali, EM, Berry, F., Heller, D., & Diguet, J.-P. (2020). Hardware-in-the-loop simulation with dynamic partial reconfiguration of the FPGA applied to computer vision in a ROS-based drone. Rapid Systems Prototyping (RSP) International Workshop 2020, 1–7. https://doi.org/10.1109/RSP51120.200.9244863

[3] D. Baumann, F. Mager, U. Wetzker, L. Thiele, M. Zimmerling, and S. Trimpe, “Wireless Control for Smart Manufacturing: Recent Approaches and Open Challenges”, in IEEE Proceedings, vol. 109, no. 4, p. 441-467, April 2021, doi: https://doi.org/10.1109/JPROC.2020.3032633 Title anhand dieser DOI in Citavi-Projekt übernehmen


Source link

Leave A Reply

Your email address will not be published.