[Paper Review] Dealing with Run-Time Variability in Service Robotics: Towards a DSL for Non-Functional Properties

Juan F. Inglés-Romero, Alex Lotz+2 more|arXiv (Cornell University)|Jan 1, 2012

Advanced Software Engineering Methodologies21 references24 citations

TL;DR

This paper proposes a Domain-Specific Language (VML) for modeling run-time variability in service robotics, focusing on non-functional properties like safety, efficiency, and energy consumption. It enables designers to specify configurable parameters (variation points) and adaptation rules at design time, which the robot dynamically resolves at run-time using constraint solving, significantly improving task execution quality in open-ended environments without pre-emptive enumeration of all scenarios.

ABSTRACT

Service robots act in open-ended, natural environments. Therefore, due to combinatorial explosion of potential situations, it is not possible to foresee all eventualities in advance during robot design. In addition, due to limited resources on a mobile robot, it is not feasible to plan any action on demand. Hence, it is necessary to provide a mechanism to express variability at design-time that can be efficiently resolved on the robot at run-time based on the then available information. In this paper, we introduce a DSL to express run- time variability focused on the execution quality of the robot (in terms of non-functional properties like safety and task efficiency) under changing situations and limited resources. We underpin the applicability of our approach by an example integrated into an overall robotics architecture.

Motivation & Objective

• To address the challenge of managing run-time variability in service robots operating in open-ended, unpredictable environments.
• To reduce design-time complexity by decoupling variability specification from immediate configuration, avoiding combinatorial explosion of pre-defined scenarios.
• To enable dynamic, context-aware adaptation of robot behavior based on non-functional properties such as safety, energy use, and task efficiency.
• To provide a mechanism for separating concerns between task planning and quality-of-service optimization through a formalized modeling language.

Proposed method

• The Variability Modeling Language (VML) is defined using an EBNF grammar to specify types, variables, context, variation points, and adaptation rules.
• Non-functional properties (e.g., safety, energy, performance) are modeled as functions of variation points, with values expressed via mathematical expressions and units.
• Adaptation rules are defined using invariants and implications to constrain variation point bindings based on current context and system state.
• VML models are translated into MiniZinc, a constraint modeling language, for execution by a constraint solver at run-time.
• The approach integrates into a layered robotics architecture, where the constraint solver interacts with the robot’s sequencing layer to bind variation points dynamically.
• A real-world coffee delivery scenario demonstrates the method, with velocity as a variation point optimized across safety, energy, and delivery time constraints.

Experimental results

Research questions

• RQ1 How can run-time variability in service robotics be modeled at design time to support dynamic adaptation without exhaustive pre-configuration?
• RQ2 What formal language constructs are needed to express non-functional property trade-offs (e.g., safety vs. efficiency) in a way that enables efficient run-time resolution?
• RQ3 How can the separation of concerns between task execution and quality-of-service optimization be achieved in a modular and maintainable way?
• RQ4 To what extent can a DSL for variability improve system robustness and performance in open-ended robotic environments?

Key findings

• The VML language successfully enables designers to express complex, context-dependent adaptation strategies for non-functional properties such as safety, energy consumption, and task efficiency.
• By deferring binding of variation points to run-time, the approach reduces design-time complexity while maintaining high system robustness in unpredictable environments.
• The integration of VML with MiniZinc and a constraint solver allows efficient, real-time resolution of optimal configurations under dynamic constraints.
• In the coffee delivery case study, the robot dynamically selected velocity based on battery level, distance, and desired coffee temperature, achieving a balance between speed, safety, and energy use.
• The method demonstrated feasibility in a real-world robotic system (robot Kate), showing that non-functional property optimization can be achieved without prior enumeration of all scenarios.
• The approach supports extensibility and validation through structured modeling, with unit specification and normalization improving model clarity and correctness.