Robotics Software: The Future Should Be Open

摘要

This column introduces a number of problem claims about the pitiful state of practice in software for robotics (and for all kinds of engineering domains in general). It also presents solution claims, whose realization can lead to a long-term, macroeconomically optimal solution, both for the industry and academia. Key problems in robotics software in the industrial and the academic practice are a chronic lack of standardization, interoperability and reuse of software libraries, both proprietary and open source. For example, we still have not standardized the Kalman filter or particle filter that everyone can and wants to use, and the same holds for many other mature robotics software components such as kinematics and dynamics, control laws, or planning algorithms. As a result, thousands of (Ph.D.) person months are lost worldwide every year in reimplementing these things for the zillionth time, without any new contribution to software reuse. This pitiful state of the practice is not unique to robotics, and only a few engineering domains do it right: numerical linear algebra (starting many decades ago already); the World Wide Web [with (X)HTML, cascading style sheets (CSS), scalable vector graphics (SVG), and other W3C standards as the fundamental enablers]; the Java middleware ecosystem [XML processing, open services gateway initiative (OSGi, now obsolete), Eclipse, mobile phone frameworks, etc.]; and tools around the Object Management Group (OMG) standards of UML, SysML, and modeldriven architecture. These examples are not tied to specific applications (this is not a coincidence but a very wise design decision about modularity and decoupling!), and they all have healthy commercial and open-source offerings, with real and rapid innovation taking place in both software development models. Every section in this article focuses on one of the fundamental issues that has led to the retarded state of software in robotics and suggests a concrete solution. Most neighboring scientific and technologic domains (computer vision, systems and control, cognitive science, artificial intelligence, etc.) suffer from exactly the same problems, such that cooperation with those domains can lead to faster implementation of the presented solutions.