- Finde uns auf Facebook
- Finde uns auf Youtube
Behaviour-based architectures came up in the 1980s. Their component-based, distributed nature paired with the ability to create large networks by combining numerous rather simple behaviours yields several advantages over classic, typically monolithic systems.
For example, a single element of a behaviour-based system can easily be developed, implemented, and tested on its own before being integrated into the target system. Furthermore, a behaviour can easily be used in different systems, which facilitates the reuse of functionality. This is usually more difficult or even impossible in monolithic architectures.
Because of the many advantages, behaviour-based components are used extensively in the control systems of the vehicles developed at the Robotics Research Lab (RRLab).
The Behaviour-based Architecture iB2C
With the aim to combine the best elements of existing architectures while overcoming their deficiencies, the behaviour-based architecture iB2C (integrated Behaviour-Based Control) is developed at the RRLab.
The iB2C has been implemented using the robotics framework MCA2-Kl (Modular Controller Architecture 2 - Kaiserslautern Branch), which is used to control all of the lab's robots.
The central component of the iB2C is a behaviour. All behaviours share a common interface consisting of stimulation s (used for gradual enabling), inhibition i (used for gradual disabling), activity a (the degree of influence the behaviour intends to have), and target rating r (the behaviour's satisfaction with the current situation). s and i are combined to the activation ι, which defines the behaviour's maximum influence within a network. These values are limited to [0,1].
Apart from the common interface, a behaviour can have special control inputs and outputs (e and u), which are connected by the behaviour-specific transfer function F: u = F (e,ι). There is no limitation on F, i.e. it can realise anything from a simple mapping to a complex calculation.
The iB2C features a special fusion behaviour that can be used to merge the outputs of multiple competing behaviours depending on their activities. Three different fusion modes are available:
- Maximum fusion: The control value of the most active behavior is forwarded.
- Weighted average fusion: The control values of the competing behaviors are weighted with the activity of the corresponding behavior.
- Weighted sum fusion: The control values of the competing behaviours are summed up according to their activity.
Extensions for Realising Sequences
Most behaviour-based architectures lack direct support for the realisation of complex, deliberative functionalities on high navigation layers. As a result, developers of high-level navigation systems cannot beneﬁt from the numerous advantages of behaviour-based approaches. However, the limitation to mainly reactive tasks is unnecessary.
Therefore, a special coordinating behaviour, the conditional behaviour stimulator (CBS), has been added to the iB2C that can be used to realise sequences of behaviour activations. The CBS' activity depends on the values at a set of special ports. To these ports, activity or target rating outputs of other behaviours are connected.Hence, a CBS gets active or inactive depending on the activities or target ratings of connected behaviours. By cascading CBSes, arbitrarily complex behaviour activity sequences can be created.
Behaviour Network Analysis
Due to the distributed nature of behaviour-based systems, the proper connection of behaviours within a network is crucial for the correct operation of a system. Therefore, the analysis and verification of behaviour networks is currently investigated at the RRLab.
A first work deals with the detection and tracing of behaviour activity oscillations. It is described in Wilhelm09. Current research deals with the modelling of behaviour networks as finite-state automata and the subsequent verification using model checking.
Formal Verification of Behaviour Networks Including Sensor Failures
in Robotics and Autonomous Systems, vol. 74, Part B, pp. 331-339, 2015, December, DOI:10.1016/j.robot.2015.08.002
Design and Verification of Behaviour-Based Systems Realising Task Sequences
Verlag Dr. Hut, München, 2015, http://www.dr.hut-verlag.de/978-3-8439-2261-6.html ISBN-13: 978-3-8439-2261-6
Soft Robot Control with a Behaviour-Based Architecture
Soft Robotics - Transferring Theory to Application, Springer-Verlag, pp. 81--91, 2015
Using Behaviour Activity Sequences for Motion Generation and Situation Recognition
Proceedings of the 8th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2011), SciTePress - Science and Technology Publications, pp. 120-127, Institute for Systems and Technologies of Information, Control and Communication (INSTICC), 2011, July 28-31, Noordwijkerhout, The Netherlands
Behaviour-Based Off-Road Robot Navigation
in KI - Künstliche Intelligenz, Springer Berlin / Heidelberg, vol. 25, no. 2, pp. 155-160, 2011, May, this publication is available at http://dx.doi.org/10.1007/s13218-011-0090-2
A Behaviour-based Integration of Fully Autonomous, Semi-autonomous and Tele-operated Control Modes for an Off-road Robot
Proceedings of the 2nd IFAC Symposium on Telematics Applications, IFAC, 2010, October 5-8, Politehnica University, Timisoara, Romania, invited paper
Development of Complex Robotic Systems Using the Behavior-Based Control Architecture iB2C
in Robotics and Autonomous Systems, vol. 58, no. 1, pp. 46--67, 2010, January, doi:10.1016/j.robot.2009.07.027
A Systematic Testing Approach for Autonomous Mobile Robots Using Domain-Specific Languages
KI 2010: Advances in Artificial Intelligence, Springer Berlin / Heidelberg, vol. 6359, pp. 317-324, 2010
Development Process for Complex Behavior-Based Robot Control Systems
Verlag Dr. Hut, 2010, http://www.dr.hut-verlag.de/978-3-86853-626-3.html
A Behaviour Network Concept for Controlling Walking Machines
Adaptive Motion of Animals and Machines, Springer Verlag, pp. 237-244, 2006