[Paper Review] Towards Robot-independent Manipulation Behavior Description

Malte Wirkus|arXiv (Cornell University)|Dec 10, 2014

Robot Manipulation and Learning8 references3 citations

TL;DR

This paper presents a robot-agnostic framework for describing and controlling manipulation behaviors using an embedded domain-specific language (eDSL) within the Rock software framework. By decoupling task-level motion planning from robot morphology through abstract control networks and reference frame abstraction, the approach enables reuse of behavior descriptions across diverse robotic systems, with first results demonstrating a modular, hierarchical control network for multi-stage manipulation tasks.

ABSTRACT

In this paper we present a workflow to design and control robot manipulation behavior. To remain independent from particular robot hardware and an explicit area of application, an embedded domain specific language (eDSL) is used to describe the particular robot and a controller network that drives the robot. We make use of a) a component-based software framework, b) model-based algorithms for motion- and sensor processing representations, c) an abstract model of the control system, and d) a plan management software, to describe a sequence of software component networks that generate the desired robot behavior. As first results, we present an eDSL for the description of a robotic system composed of mechatronic subsystems, and for the creation of a multi-stage control network.

Motivation & Objective

• To enable robot-agnostic description of manipulation behaviors independent of hardware morphology.
• To decouple task-level motion planning from robot kinematics and sensor data processing.
• To develop a reusable, composable control system using component-based software and abstract control models.
• To support behavior composition and transfer across different robotic platforms through a unified control network abstraction.
• To enable efficient integration of sensor processing, motion planning, and low-level control via a domain-specific language.

Proposed method

• The framework uses a component-based software architecture built on the Rock framework to modularize robot control components.
• It introduces an embedded domain-specific language (eDSL) to define control networks and hierarchical controller structures.
• Motion plans are abstracted as parametric functions (e.g., DMPs, waypoint navigation) independent of robot kinematics.
• Sensor data is processed through pipelines that transform raw input into Cartesian-space references in sensor-specific frames.
• Control signals are managed via a priority-based cascade controller architecture that resolves conflicts between overlapping control commands.
• Transformations between reference frames (sensor, task, robot, joint) are handled through a standardized geometric description convention.

Experimental results

Research questions

• RQ1 How can manipulation behaviors be described independently of robot morphology and hardware?
• RQ2 What abstraction mechanisms enable reuse of behavior descriptions across different robotic platforms?
• RQ3 How can sensor processing, motion planning, and low-level control be integrated into a unified, composable control network?
• RQ4 What role does a domain-specific language play in simplifying the specification of complex, multi-stage control systems?
• RQ5 How can conflicting control signals from multiple subsystems be resolved in a hierarchical control architecture?

Key findings

• An embedded domain-specific language (eDSL) was successfully implemented to define multi-stage control networks for manipulation tasks.
• The framework enables the creation of hierarchical control systems where high-level motion plans are decoupled from robot-specific kinematics.
• A cascade controller architecture allows for modular composition of control signals, with priority-based conflict resolution between controllers.
• The system supports virtual control interfaces that abstract joint-level control into higher-level functional units (e.g., 'cart_arm_plus_wrist', 'head', 'finger').
• The eDSL is extensible and enables the integration of complex component networks using Syskit compositions and dataflow interfaces.
• The approach demonstrates feasibility for transferring behavior descriptions across robots by abstracting control logic from physical morphology.