Skip to main content
European Commission logo print header

A computing toolkit for building efficient autonomous applications leveraging humanistic intelligence

Periodic Reporting for period 1 - TEACHING (A computing toolkit for building efficient autonomous applications leveraging humanistic intelligence)

Período documentado: 2020-01-01 hasta 2021-06-30

The world is on the verge of the autonomous systems revolution. Automation is the technology enabling the conduction of processes with minimum human assistance and can be used to operate Cyber-physical Systems of Systems (CPSoS) comprising complex, multi-faceted and dynamic virtual and physical resources. However, even when the most advanced degree of autonomy is exercised, the human is a variable which cannot and should not be left out of the CPSoS equation, especially in safety critical scenarios. Humans interact with the autonomous system either as passive end-users or as active co-operators in a mutual empowerment relationship towards a shared goal. Considering autonomous CPSoS from a human-aware perspective brings-in critical requirements in terms of adaptivity, dependability and privacy. Not to mention, a careful consideration of human comfort and distress throughout system operation. Nevertheless, it also enables unparalleled innovation potential throughout the realization of a holistic environment where the human and the cyber-physical entities support, cooperate and empower each other.
Artificial Intelligence (AI) is a key technology to realize autonomous applications, even more so within the interacting context of a CPSoS. The stringent computational and memory requirements of AI will impose a significant rethinking of the underlying computing software and system which will need to provide AI-specialized support in the computing fabric, even at a hardware level. The realization of such an intelligent empowerment of the CPSoS will also require addressing challenges related to AI fundamentals as well as to dependability issues in distributed intelligent and autonomous systems
The H2020 TEACHING project stems from the need of providing an answer to the following compelling research questions:
- How to construct a cooperative human-CPSoS environment placing the needs, the comfort and the well-being of the human at the core of the CPSoS operation?
- How can such a cooperative environment be realized to operate in an autonomous, safe and dependable way, while being capable of self-adapting by exploiting sustainable human feedback?
- How to change the underlying computing system, at an architectural and software level, to support the operation of such an adaptive, dependable and human-centric CPSoS?
Providing an effective answer to these questions is fundamental for a safe diffusion of autonomous AI-enabled applications in the European society. This is of particular relevance for many safety-critical applications, such as in automotive, avionics and general autonomous transportation In order to address this challenge, TEACHING aims to develop a human-aware CPSoS for autonomous safety-critical applications, based on a distributed, energy-efficient and dependable AI, leveraging innovative edge computing platforms integrating specialized computing support for AI and dependability. The goal of the TEACHING project is to design a computing platform and the associated software toolkit supporting the development and deployment of autonomous, adaptive and dependable CPSoS applications, allowing them to exploit a sustainable human feedback to drive, optimize and personalize the provisioning of their services. The project achievements will be demonstrated in an automotive and an avionic industrial use case, which are highly relevant for the European societal and industrial ecosystem, and that pose high challenges when it comes to dependable interactions between a system operating an intelligent task and the human.
The Consortium dedicated to the design of the overall TEACHING system and its components. It has been defined a high-level architecture of the TEACHING software, composed of five toolkits implementing the key functionalities. On the system level, the Consortium completed the definition of the supported computing and communication infrastructure. It has been identified an heterogenous set of devices supported by the TEACHING CPSoS, including the design of a real-time embedded platform providing specialized hardware support both for running AI-models as well as cybersecurity. The technical work packages led to the development and release of a first integrated mock-up of the TEACHING system. This includes the AI-as-a-service (AIaaS) toolkit which provides the implementation of neural network models designed to scale from embedded-systems to cloud execution, and prototype learning mechanism for continual, federated and dependable learning. A prototype model for human-centric personalization has been implemented leveraging reinforcement learning and human stress state estimation methods. The execution of such learning models is orchestrated by the HPC2I toolkit providing initial abstractions for computing and communication leveraging containerization and publisher-subscriber frameworks. The Windflow library for stream-like computing on Edge/IoT resources has also been publicly released. The project also designed three engineering patterns to ensure dependability of autonomous applications running AI-based components, showing their impact in small-scale automotive scenarios. The focus of the use case implementation activities has been primarily on the adaptation of the technology bricks to allow integration in the industrial pilots, including the porting of technologies available at the industrial partner sites on the TEACHING embedded real-time platform. The TEACHING research effort has already led to the publication of 20 scientific papers.
A key outcome of TEACHING will be the creation of a concept of an integrated human-software-system environment where the human and the cybernetic entities collaborate synergistically, where the latter provide the former with a comfortable, tailored and dependable interaction driven by the implicit feedback provided by the human throughout his/her physiological reactions to CPSoS operation.
Apart from the obvious societal implications of a cybernetic system that is respectful and adaptive to the individuals’ reactions, the project outcomes will have heavy impact at an industrial level. Considering the project uses cases, the autonomous transportation industries might leverage TEACHING results to build systems that operate safely while taking into consideration the physiological and emotional responses of the humans involved. In the aviation domain, TEACHING will provide the tools to build intelligent cyber-blackboxes that can monitor the vessels, anticipate and detect issues and promptly propose solutions and mitigations in interaction with the pilot. Impacts of the project are far deeper than use cases alone and involve environments where cyber-autonomous applications operate in close physical interaction with the human, such as in assembly lines. Key to such goals is the progress in challenging research objectives which the project tackles and promotes under the umbrella term of Pervasive AI. These include (i) introduce the use of dependability engineering methods for trustworthy AI, (ii) developing sustainable continual learning specific for stream data, (iii) design new learning algorithms leveraging imprecise feedback, such us human reactions, and (iv) develop computing and communication abstractions to ease distributed deployment and energy-efficient execution of AI-based applications.
The TEACHING integrated human-software-system concept