Software development for robotic systems require knowledge from several domains. This paper highlights the application of domain modeling in SafeRobots Framework for designing robotic systems. We discuss how heterogeneous domain models can be modeled and integrated for systematic software development, and how it can be employed in intelligent model evolution.
System engineering robots poses an interesting challenge of integrating
heterogeneous problem domains while designing the system. This heterogeneity can be
seen in terms of conceptual domains (e.g., perception, control systems, mapping),
models of computation (e.g., synchronous, asynchronous, discrete, continuous
time), application domains (e.g., service robots, industrial robots, driver-less
cars), etc. In the last decade, developing software for such complex systems was
addressed by `divide and rule' strategy by adopting component-based software
engineering techniques [
In this paper, we highlight the application of domain modeling for robotic
system design in our `SafeRobots' framework [
By learning from the shortcomings of code-based approaches, the software
engineering community in robotics is gradually moving towards Model-Driven
Software Development (MDSD) approach. The Smartsoft framework is based on
a model driven toolchain that support formal modeling of component
skeleton that act as a wrapper around the user code [
SafeRobots: A model-driven toolchain for software development in Robotics
Self Adaptive Framework for Robotic Systems (SafeRobots) is a model-driven
toolchain that is currently under development in our lab at ENSTA-ParisTech.
SafeRobots is based on three software engineering paradigms: Knowledge-based
engineering, Model-driven engineering, and Component-based software
engineering. A comprehensive discussion on SafeRobots can be found in [
In SafeRobots framework, the entire software development process is conceptually divided into three spaces: problem space, solution space, and operational space. Each space is supported by a knowledge space which acts as a bridge between the three spaces. The complete ecosystem is illustrated in Figure 1. In problem space, the problem, the requirements, and the contexts are modeled using appropriate Modeling Languages (ML). The solution space model captures the knowledge relating to domain concepts, computational algorithms, execution sequence, and their non-functional properties in a formal way. The Architecture Modeling and Analysis Language (AMAL) and its open semantic framework enables concrete architecture modeling in the operational space. 4
In SafeRobots, the domain knowledge is modeled in two di erent phases: Problem Independent Knowledge Modeling and Problem Speci c Knowledge Modeling. These two phases correspond to the knowledge space and solution space in Figure 1. 4.1
The domain knowledge modeling in this phase is independent of the problem speci cation or application constraints. The domain concepts are formally modeled using ontologies, Domain-Speci c Languages (DSLs), Knowledge graphs, etc. The models at this level captures the robotic domain speci c concepts, meta-data about the computational algorithms and standard interfaces, their structural dependencies, etc. The domain knowledge complements the various application speci c development process by providing a knowledge base for abstract concepts such as image, point clouds, links, joints, platform, etc. 4.2
Problem-speci c knowledge modeling or solution space modeling is performed
with the help of functional requirements from the problem model as constraints
applied to the domain model. In other words, a solution space model captures
multiple solutions for the given problem by considering only the functional
requirements and given domain knowledge base modeled in problem independent
knowledge modeling phase. The strategy is to postpone the decisions on
nonfunctional requirements at a later stage, since such properties can be estimated
only when platform-speci c decisions are made. In our approach, the solution
space is formally modeled using our modeling language, `Solution Space
Modeling language (SSML)'. Figure 2 shows a solution space model in SSML for lidar
based vehicle tracking application. The connectors represents computational
algorithms and their non-functional property model and the gates represents basic
operations for composing di erent functional computational processes. The given
model shown in Figure 2b capture three di erent solutions for the pointcloud
segmentation problem. The three solutions that are modeled satis es the
functional goal of the problem model, but has di erent non-functional properties. A
detailed discussion on SSML and the vehicle tracking application scenario can
be found in [
The conceptual spaces from the SafeRobots framework shown in Figure 1 are
hierarchically arranged in such a way that the lower layer uses the knowledge
gained in the upper layer. The knowledge space consists of domain concepts
represented using ontologies. The problem is modeling in the form of goals (i.e.,
hard goals and soft goals) and requirements modeled using Goal and
Requirement Language (GRL). By applying the functional constraints provided by the
problem space on the domain conceptual knowledge, the solution space for the
given problem is modeled. The non-functional properties (NFP) modeled using
our NFP modeling language [
The SafeRobots toolchain is being implemented using Eclipse Modeling Framework. The model evolves during the software development by incorporating the knowledge created in the form of models and helps to shift the critical decision when more information is available.
In this paper, we discussed how modeling techniques are employed in SafeRobots framework to model domain knowledge and how they can be applied in intelligent software tools and process, to develop complex robotic systems. The main challenge is to adopt the domain model at the appropriate granularity to assist the system designer in systematic software development process to develop e cient and reusable software for robotic systems.